YC Summer 2026 Batch, Ranked. The B2B Startups Worth a First Meeting
All 122 B2B companies in the YC Summer 2026 batch, scored with quant models and two taste profiles. Built by Vela, a tier-1 quant VC.
Yigit Ihlamur
Co-founder, General Partner
September 1, 2026·
The YC S26 batch has 236 launched companies and almost no public data on any of them. We researched the founders on LinkedIn, ran every company through the prediction models of Vela, a tier-1 quant VC, scored them against two taste profiles, Vela's own and one built from the public track record of a16z, a tier-1 VC, and mapped their competitors. This page covers all 122 B2B companies in the directory and is built for one job. Filter, sort, and decide who to talk to first.
Every demo day, thousands of investors triage the same batch with the same three tools. A directory page, a launch post, and whatever a founder's LinkedIn happens to say. We think the ecosystem works better when that homework is done once, in the open, with data anyone can inspect. So we are publishing ours. This is what a quant VC's homework looks like when it is done in the open. Each company below carries verified founder backgrounds, output from Vela's founder prediction model, an estimated ARR where our estimator had enough signal, the verdicts of the Vela and a16z taste models, and funded competitors for context. Founders, if anything here is wrong or out of date, tell us and we will fix it.
01
The batch at a glance
Before the table, the shape of it. Every count in this section except the thesis clusters is computed live from the dataset below, so it stays true as rows get corrected.
122 B2B companies in the batch
27solo founders
23disclose live revenue
31live deployments
54teams above 5x baseline
9Vela taste engages
Most of these companies are weeks old. Read every aggregate as a reading of teams and early evidence, not of businesses.
Where the two taste models land
a16z Engage
a16z In-between
a16z Pass
Vela Engage
7
2
0
Vela In-between
36
11
13
Vela Pass
12
11
30
Consensus is rare and passes are cheap. Both models engage only 7 of 122 companies, the closest thing this page has to a shortlist. The 12 companies where the a16z read engages but Vela passes are where pattern-matching outruns conviction, and where price discipline matters most.
The founder model is built to be long-tailed against a 1.9% baseline outlier rate, and the tail is where to hunt. 54 teams read above 5x, and the 16 above 10x are listed here. A high multiple is a reason for a meeting, not a verdict.
Crowded corners, six theses the batch keeps repeating
Inference economics, routing, token cost, capacity1512% of batch
Environments, evals and training data1311% of batch
Agent-run services firms1311% of batch
Coding-agent fleet tooling87% of batch
Agent memory and shared context65% of batch
Security for the agent era65% of batch
Analyst clustering, memberships recorded in the raw dataset. Six theses absorb 61 of 122companies, half the batch selling to the other half's stack. Inside a crowded corner the question in the room is not whether the thesis is right but why this team wins it, so the team read carries more weight than the demo.
The read. This batch is long on agent infrastructure and short on disclosed evidence, 23 of 122 companies show live revenue and 31 show live deployments, so most first meetings are team meetings rather than traction meetings. The two taste models agree far more on passes than on conviction, which is the honest state of day-zero investing, and the 7 double engages plus the 16 strongest founder-model reads are the two lists worth working through first. Everything below exists to argue with these aggregates one company at a time.
02
The scoreboard
All 122 B2B companies in the batch, one row each. Click a row to jump to its full profile. The rank blends four lanes, weighted 35% team read, 25% taste (the mean of the Vela and a16z reads), 20% traction evidence, and 20% Vela's founder prediction model normalized within this cohort. Every score on this page is shown relative to the collected population, computed directly from the dataset. Click any column header to sort, or change the lane in the toolbar.
122 companies
Any of the chips you pick within a row, and all of the rows together.
Founder model multiples come from Vela's founder prediction model against a 1.9% baseline outlier rate. Vela taste is the conviction profile of Vela, a tier-1 quant VC, and a16z taste is a market read modeled on the public record of a16z, a tier-1 VC, both reported unedited. Relative bands (Top 10% through Bottom 25%) are percentile ranks computed in code from this page's own dataset and update automatically as coverage grows. ARR figures are Vela OS estimates unless the company disclosed them. Traffic and funding signals come from Vela OS traction tracking.
Full inventory of every AI app, MCP and skill on every company device with real-time blocking. 2,000+ devices live across six enterprises, pilots on 10,000+.
Overall
Top 10%
Team read
Top 25%
Founder model
19.3x vs 1.9% base
ARR
-
Visits
-
Xia HuaCEODirector of Engineering at Clumio for ransomware protection. PhD Applied Math MIT.
Varun WadhwaCTOSenior SWE at LinkedIn on RAG database systems, Staff Engineer at Clumio, payments at Microsoft. Berkeley EECS.
TechnicalAbove median
ShippingTop 25%
SellingTop 25%
Domain fitAbove median
Self-awareTop 10%
FocusTop 25%
Taste readsa16z engages, Vela stays in-between.
Who they are up against
Harmonic Security$50M+ raisedAI usage visibility and data protection for enterprises.
Netskope / ZscalerincumbentsGateways that cannot see on-device agent actions.
NYC · solo founder · partner Brad Flora · founded 2024
One API for fundamentals, filings, ownership and prices across 27,000+ global stocks, built agent-first with citations. 1,000+ customers including hedge funds, millions of requests daily.
Overall
Top 10%
Team read
Top 10%
Founder model
10.9x vs 1.9% base
ARR
-
Visits
-
Virat SinghCEO90K+ GitHub stars across financial agent projects (ai-hedge-fund at 60K, Dexter at 27K). Tech lead at Faire, engineer at Airbnb and Acorns, started in buy-side M&A.
TechnicalTop 25%
ShippingTop 10%
SellingTop 25%
Domain fitTop 25%
Self-awareTop 25%
FocusTop 25%
Taste readsa16z engages on owned distribution (90K stars) and real revenue from 1,000+ customers. Vela's taste holds at in-between on solo-founder concentration.
An accounting firm with zero software to learn. Integrate, and agents close the books. Partnered with Series B to D companies representing over $1B in enterprise value within three months.
Overall
Top 10%
Team read
Top 10%
Founder model
10.7x vs 1.9% base
ARR
$406K
Visits
4.4K
Joan McIntoshCEOBuilt and exited Opaper, an accounting/ERP platform that reached 45,000 businesses in four years. Earlier a senior AI/ML product manager at StreetLight (acquired) and Lacuna, ex-Merrill Lynch, once en route to a CPA/CFA.
Philip MoniagaCTORepeat YC founder (Finnix, W22). Built financial data infrastructure at Bloomberg and Booking.com, CTO of a trading platform with $1B+ in volume, earlier senior engineering at Carousell.
TechnicalAbove median
ShippingTop 25%
SellingTop 10%
Domain fitTop 10%
Self-awareTop 25%
FocusTop 25%
Taste readsAn exited founder returning to the exact market of her exit, paired with a repeat YC CTO from Bloomberg. The taste model's shipped-artifact and narrow-depth rules both fire positive.
Who they are up against
Pilot$172M raisedThe scaled tech-enabled bookkeeping incumbent this generation is attacking with agents.
Puzzle$45M raisedAI-native accounting software, sells a tool where Billow sells the outcome.
Truewind$16.5M raisedYC-born AI accounting service, the closest like-for-like competitor.
Two frontier agents race the same real task, humans judge blind, and every battle becomes an eval no model has seen. From zero to $60K revenue in the first two weeks, users doubling weekly.
Overall
Top 10%
Team read
Top 10%
Founder model
5.4x vs 1.9% base
ARR
$60K
Visits
-
Prateek JannuCEOStanford dropout. Built the #1 agent on OSWorld at 82% accuracy, lead ML engineer for computer-use agents at TruGen, ML research at Purdue with a US provisional patent.
Nitish KovuruCTO/COOColumbia CS (vision track), Purdue computer engineering, TA'd a graduate AI course as an undergrad. Shipped Coasty, a SOTA computer-use framework, before the arena pivot.
TechnicalAbove median
ShippingTop 10%
SellingTop 10%
Domain fitTop 25%
Self-awareTop 10%
FocusAbove median
Taste readsRevenue in week two and an artifact the labs cannot build themselves. Speed to evidence is exactly what the taste model's gasoline rule looks for.
Who they are up against
LMArena~$100M seed (reported)The chatbot arena playbook this extends to computer-use, backed by a16z and UC Investments.
Braintrust$121M raisedEval infrastructure for AI products, the enterprise-tooling flank.
Scale AI$15.9B raisedThe incumbent human-eval and data pipeline for every frontier lab.
LA · team of 5 · partner Brad Flora · founded 2025
Restaurants subscribe at $399 a month and vetted local creators show up, film and post, with AI agents handling matching, booking and 10,000+ iMessages a month. 120+ restaurants, 37% MoM growth, 90%+ retention, 35M views.
Overall
Top 10%
Team read
Top 10%
Founder model
9.2x vs 1.9% base
ARR
-
Visits
-
Yousef AbdelfattahCEOCo-founded FaZe Clan after starting as a 14-year-old YouTuber, helped take it public on Nasdaq at a $1B valuation. 15 years in the creator economy, 10M+ followers.
Obaida AlbaroudiCOOEarly Replit employee, ex Goldman Sachs, co-founded Clipping Exe and grew a consumer app from $360K to $2M ARR in 18 months.
Ahmad IbrahimCTOCo-founded Clipping Exe, infrastructure tracking 10B+ views. Berkeley CS.
TechnicalBelow median
ShippingTop 10%
SellingTop 10%
Domain fitTop 10%
Self-awareTop 25%
FocusTop 25%
Taste readsa16z engages on verified revenue, retention, and an IPO-proven founder. Vela's taste rejects the services-network shape, creators and restaurants rather than compounding software.
Who they are up against
Aspire~$36M raisedInfluencer marketing platform for brands.
YelppublicThe legacy local discovery incumbent.
Ad agenciesn/aThe thousands-a-month status quo TryNearby undercuts at $399.
Nyx continuously red-teams customer-facing AI through its environment, no source access needed, adapting attacks over 10,000+ strategies. 78% attack success on AgentHarm versus 67% for the best frontier model, exploits found at dozens of Fortune 500s.
Overall
Top 10%
Team read
Top 10%
Founder model
11.5x vs 1.9% base
ARR
-
Visits
-
Ahmed AlyCEOFirst data scientist at Sequoia-backed Two, built its fraud engine to $1B+ in annual transactions. Led international payments fraud at Monzo, published in Q1 journals, UCL PhD dropout.
Ibrahim AbduCTORed-teamed AI agents in Meta's privacy org, built an SRE agent triaging 3M+ errors a month, compilers for insider-trading detection at TradingHub. Oxford PPE, top 8%. Worked with Ahmed before.
TechnicalTop 25%
ShippingAbove median
SellingAbove median
Domain fitTop 10%
Self-awareTop 25%
FocusTop 25%
Taste readsa16z engages, red-teaming is squarely the control plane it funds and the benchmark beats frontier models. Vela's taste holds at in-between until Fortune 500 findings convert to contracts.
Who they are up against
Lakera~$30M raisedAI security and guardrails, the funded adjacent.
Haize LabsseedAutomated red-teaming research lab, the closest thesis rival.
HackerOne$159M raisedThe human red-team marketplace incumbent.
Agents monitor legal-specific data sources to surface clients and cross-sells; partners log in daily to see who to contact and why now. $0 to ~$15K MRR in 4 weeks.
Overall
Top 10%
Team read
Above median
Founder model
12.1x vs 1.9% base
ARR
-
Visits
-
Jity WoldemichaelFounderMIT CS, engineering at Meta and Jane Street.
Kali AbejeFounderMIT CS, public policy minor, NLP research with MIT GOV/LAB, 2x AWS intern, debate and mock trial.
TechnicalBelow median
ShippingTop 25%
SellingTop 25%
Domain fitBelow median
Self-awareBelow median
FocusAbove median
Taste readsa16z engages, Vela stays in-between.
Who they are up against
IntapppublicLaw firm business development software incumbent.
Harvey$500M+ raisedCould bundle BD once it owns the firm's work product.
Seattle · team of 2 · partner Brad Flora · $1.1M pre-seed Jun 2026
Routes workloads to the best available accelerator across NVIDIA, TPU, Trainium and AMD in 4+ clouds, and lets GPU owners plug capacity into the rails. Live at 100 billion tokens a week across 22 locations.
Overall
Top 10%
Team read
Top 10%
Founder model
5.7x vs 1.9% base
ARR
$185K
Visits
4.8K
Jaden WangCEOBuilt a half-megawatt data center from a warehouse shell at 20, ran virtualization and HPC at Voltage Park, ex TensorDock, UW dropout, founded Heaviside Compute.
Prashant PatelCTOStaff engineer at Voltage Park running 10,000+ GPU clusters, founding member of Amazon Bedrock where he took Custom Model Import from zero to multi-million ARR. NYU MS CS.
TechnicalTop 10%
ShippingTop 10%
SellingAbove median
Domain fitTop 10%
Self-awareTop 25%
FocusTop 25%
Taste readsA live inference delivery network at 100B tokens a week reads as an open artifact door, and the founders own the exact function from Voltage Park and Amazon Bedrock.
Who they are up against
Together AI$1.33B raisedScaled open-model inference cloud.
SF · team of 4 · partner Harshita Arora · $1M pre-seed Jun 2026
Executes source-to-contract procurement inside customers' own systems. $400M a year of supplier spend under management, multiple six-figure contracts, and a $1M pre-seed, all before demo day.
Overall
Top 10%
Team read
Above median
Founder model
6.6x vs 1.9% base
ARR
$204K
Visits
3.3K
Emre IsikCEOCambridge computer engineering, dropped the master's. Researched motor-control algorithms at UCLA's IPAM, trained with Turkey's informatics olympiad reserve team at 15.
Hakan Enes AksuCTOBogazici dropout shipping commercial software since 16, $40K Reddit ad-automation service, sold a tender-matching engine for five figures, top 500 of ~3M in Turkey's national exam.
Ali Eren AytekinCo-founderITU dropout, Microsoft Imagine Cup EMEA runner-up for a 98.3% genetic-signal classifier, holds a compressed-air energy storage patent.
TechnicalBottom 25%
ShippingAbove median
SellingTop 10%
Domain fitBelow median
Self-awareTop 25%
FocusTop 25%
Taste readsReal spend under management and six-figure contracts within months, an AI-workforce wedge into procurement priced against labor rather than seats.
Two stock coding agents plus hiloop's infrastructure ran 4,188 experiments in two days and beat the published state of the art on Karpathy's autoresearch benchmark. Now pointed at customer post-training and optimization problems.
Overall
Top 25%
Team read
Top 25%
Founder model
14.3x vs 1.9% base
ARR
-
Visits
-
Karan BrarCEOML at Reducto, Head of ML Infrastructure at DynamoAI (2 years), three years at BMO before that. The infra half of the pair.
Thomas BoserCTOEngineer at Reducto, founding engineer at Crosswise, senior ML engineer at Discord, earlier ML at Sentropy and SoFi. A serial early-stage builder.
TechnicalTop 25%
ShippingTop 25%
SellingBelow median
Domain fitTop 25%
Self-awareTop 25%
FocusAbove median
Taste readsBeat published SOTA with a live artifact and both founders paired from Reducto. The taste model reads it as control-plane infrastructure with real velocity.
Who they are up against
Sakana AI~$244M raisedFrontier lab built around AI-automated research discovery.
Boston · team of 5 · partner Jared Friedman · founded 2025
A deterministic tax engine behind a single API, IRS-authorized as software developer, transmitter and intermediate service provider. Federal 1040 e-file approval landed in August, covered by trade press.
Overall
Top 25%
Team read
Top 25%
Founder model
13.4x vs 1.9% base
ARR
-
Visits
-
Angelo PolicicchioCEOCPA. Forensic accounting with the FBI, equity research at Credit Suisse, public accounting at EY, Harvard MBA. Spent years inside the legacy tax software he is replacing.
Akash SadashivapethCTOEmployee #1 and head of data science at Welligence Energy Analytics. Penn data science, Harvard MBA.
TechnicalBelow median
ShippingAbove median
SellingTop 25%
Domain fitTop 10%
Self-awareTop 25%
FocusTop 25%
Taste readsa16z engages on the IRS-authorization artifact and the cohort's top founder-model score. Vela's taste rejects, reading the CPA-services lane and licensing burden as outside its thesis. The widest taste split in the batch.
Who they are up against
IntuitpublicThe consumer and pro tax incumbent.
april$78M raisedEmbedded tax engine for fintechs, the closest funded rival.
Keep using GPT or Claude. Their platform builds a simulation of your workload from traces, trains a model you own against it, and guarantees at least 50% lower cost at equal or better quality.
Overall
Top 25%
Team read
Top 10%
Founder model
8.5x vs 1.9% base
ARR
-
Visits
5.2K
Kion FallahCEOML PhD from Georgia Tech on synthetic data. Rose from intern to Staff Research Scientist at Waabi in 2.5 years, leading the mixed-reality simulation team for autonomous vehicles. South Park Commons EIR before founding.
Silen NaihinCTOFounding AI engineer of AutoGPT, grew it to ~160K GitHub stars and built one of the first agentic benchmarks (NeurIPS SoLaR paper). Co-founded Stackwise (YC W24), DOE Genesis technical staff, ICML 2026 spotlight on interpretability.
TechnicalTop 10%
ShippingTop 25%
SellingBelow median
Domain fitTop 25%
Self-awareTop 10%
FocusTop 25%
Taste readsTwo researchers who own the standard they built, in a narrow infrastructure layer between the model and the application. Precisely the profile the taste model's depth rules reward.
Who they are up against
OpenPipe$6.7M raised, acq. by CoreWeaveFine-tuning from production traces, validated the motion and exited early.
Predibase$28.5M raised, acq. by RubrikSmall-model training and serving platform.
OpenRouter$153M raisedThe closed gateway their open-source wedge is aimed at.
An AI-run collections agency negotiating with 100K+ borrowers weekly, claiming wins in 100% of head-to-head pilots.
Overall
Top 25%
Team read
Top 25%
Founder model
8.3x vs 1.9% base
ARR
$258,973 est
Visits
-
Diego AvellanedaCEOApplied ML researcher, two peer-reviewed publications, 6+ years embedded ML, built Audi's e-commerce platform at CI&T. Second-time founder.
Daniel MartinezProductCPO at Treble (YC S19), 5+ years leading B2B products.
TechnicalBelow median
ShippingTop 10%
SellingTop 25%
Domain fitAbove median
Self-awareBelow median
FocusTop 10%
Taste readsThe widest kind of split, a16z engages while Vela passes. Founder model reads 8.3x.
Who they are up against
Salient$60M+ raisedAI voice agents for auto-loan servicing and collections.
Local entrepreneurs run wearable-camera data businesses capturing how physical work is done; multiple million-dollar contracts with frontier labs signed.
Wave opens multi-gigabyte waveform files before the incumbent viewer finishes its license check, and its agent traces a failing signal to root cause. Picking a fight with the $20B EDA tooling oligopoly.
Overall
Top 25%
Team read
Top 10%
Founder model
15.5x vs 1.9% base
ARR
-
Visits
-
Athreya AnandCEOTech lead at Google for 3.7 years across agentic workflow tools and prediction models for Google Ventures. Georgia Tech MS in ML, earlier Amazon and Tesla internships.
Rishov SarkarCTOGeorgia Tech ECE PhD on chip simulation and compilers, 2022 ECE Graduate TA of the Year. Built simulators for AMD's AI Engine accelerator and LLVM compilation flows at Siemens EDA.
TechnicalTop 10%
ShippingAbove median
SellingBelow median
Domain fitTop 10%
Self-awareTop 10%
FocusTop 10%
Taste readsA PhD whose thesis is the product, a Google tech lead who left to join him, and a decade of friendship behind the pairing. Shipped tools against incumbents whose users visibly hate them.
Who they are up against
Synopsyspublic, ~$85B capOwns Verdi, the industry-standard debugger Wave undercuts.
Cadencepublic, ~$90B capThe other half of the EDA duopoly.
ChipStack~$25M reportedAI-native chip verification startup, the venture-backed flank.
Agents post progress, blockers and decisions into one chat across Claude Code, Codex and OpenCode, and engineers steer from there. 410 engineers on it, 60+ paying, 1.2 hours a day average use.
Overall
Top 25%
Team read
Top 10%
Founder model
7.4x vs 1.9% base
ARR
-
Visits
-
Aniket DeshpandeCEOApplied AI at Meta Superintelligence Labs. Built agent orchestration, RL and search infra, grew Meta AI MAU by 250K, fed the Llama training data flywheel.
Mugdhaa PatankarCTOLed AI developer productivity at Snowflake, before that live video distribution at Twitch serving 10M+ concurrent viewers, and AWS. Friends with Aniket for nearly a decade.
TechnicalTop 25%
ShippingTop 25%
SellingAbove median
Domain fitAbove median
Self-awareTop 10%
FocusTop 25%
Taste readsStrong Meta and Snowflake pedigree with real paying users, held back only because the co-founder's record was not independently verifiable to the model.
Who they are up against
Cursor (Anysphere)$3.4B raisedThe agentic coding gravity well that could ship this natively.
Warp$73M raisedAgentic terminal converging on the same surface.
Conductor and OSS session managersseed stageA fast-growing pack of agent-orchestration UIs.
Each employee gets a sandboxed agent wired to their tools while admins hold one policy. Credentials are injected at the network layer after authorization, so agents never hold secrets.
Overall
Top 25%
Team read
Top 25%
Founder model
5.4x vs 1.9% base
ARR
$232K
Visits
-
Jonathan FishnerCEOBuilt zero trust network access as Principal Security Architect at Axis Security (acquired by HPE), founded ChartDB and grew it to 20K+ GitHub stars and 100K developers, IDF and Fiverr engineering before that. OneCLI itself is past 350K downloads.
Guy Ben-AharonCTOFirst employee at Argon, the AppSec company acquired by Aqua Security, protecting client codebases.
TechnicalTop 25%
ShippingTop 25%
SellingBelow median
Domain fitTop 10%
Self-awareTop 25%
FocusTop 25%
Taste readsAn open-source artifact with 350K+ downloads, 3,200+ stars and enterprise logos like Docker and Zoho, plus a CEO who previously grew ChartDB to 20K stars. Machine identity for agents is exactly the control plane the profile funds.
Who they are up against
Teleport$169.2M raisedMachine identity and infrastructure access, the scaled adjacent incumbent.
Composio$31M raisedTool access and auth layer for agents.
Zenity$59.5M raisedAgent security and governance for the enterprise.
Agents execute every non-physical front and back of house task inside the hotel's existing systems via computer use, no integrations. Trusted by Michelin Key hotels and live across four continents.
Overall
Top 25%
Team read
Above median
Founder model
6.8x vs 1.9% base
ARR
-
Visits
-
Adrian StoicaCEOGrew up in hospitality and worked both sides of the desk. AI awareness research, co-produced records totaling 40M+ streams at Sony Music.
Saman SayahpourCo-founderAdmitted to study mathematics at the University of Vienna at 14, CMU Logic and Computation dropout, research across causal ML, prediction markets and photonic computing.
TechnicalBelow median
ShippingAbove median
SellingTop 25%
Domain fitAbove median
Self-awareBelow median
FocusAbove median
Taste readsThe a16z read engages on a live artifact in Michelin Key hotels across four continents. Vela's taste holds at in-between, wanting revenue quality behind the deployments.
Who they are up against
Mews~$410M raisedThe modern hotel PMS incumbent.
Cloudbeds$248M raisedHospitality management platform at scale.
Canary Technologies$177M raisedAI guest management for hotels, the closest funded rival.
Automates fraud investigations, KYB/KYC and loan underwriting with self-evolving, always-compliant, fully-auditable workflows.
Overall
Top 25%
Team read
Top 25%
Founder model
11.8x vs 1.9% base
ARR
-
Visits
-
Naveen QureshiCEOFounded and sold Sable (YC S19) after scaling to a few hundred thousand accounts. Ex Capital One, Amazon. Columbia MBA.
Frederik VanhevelCTOFounding engineer at Sable, rebuilt Seen's infrastructure through the Snap Finance acquisition, built its credit decisioning engine.
TechnicalAbove median
ShippingAbove median
SellingAbove median
Domain fitTop 25%
Self-awareTop 10%
FocusTop 25%
Taste readsRare double engage, both taste models lean in.
Who they are up against
Greenlite$36M raisedAI compliance analysts for fintech ops teams.
Unit21 / Sardine$100M+ raised eachFraud and compliance platforms adding agents.
SF · team of 2 · partner Jared Friedman · backed by Reach Capital pre-YC
Connects to a facility's building systems, learns how the site behaves, simulates fixes before touching anything, and coordinates the work. Live in 34 facilities with $400K+ of avoidable energy cost identified.
Overall
Top 25%
Team read
Above median
Founder model
0.8x vs 1.9% base
ARR
$364K
Visits
7.2K
Hursh ShahCEO19. Started Edviro at 17 after finding $360K of gas overbilling in his school district's utility bills. OpenAI ChatGPT Lab (1 of 30 students), NeurIPS 2024, previously founded Skyglass (PearX). Top 0.6% chess player.
Tanuj SiripurapuCTO19, programming since age 9. RISC-V RF signal classification at RTX, scaled a digital agency in high school, co-founded MediHacks, the largest global health hackathon series.
TechnicalBottom 25%
ShippingTop 25%
SellingTop 25%
Domain fitBelow median
Self-awareTop 10%
FocusTop 25%
Taste readsA live artifact in 34 facilities beats any resume. The taste model's shipped-artifact rule dominates, and the age reads as speed rather than risk once the deployments are real.
Who they are up against
Phaidra$92.5M raisedDeepMind-alumni AI control for industrial plants and data center cooling.
Runwise$95M raisedBuilding operations platform across 10,000+ buildings.
BrainBox AI$75M raisedAutonomous HVAC optimization, the closest energy-AI incumbent.
A 10-minute voice interview becomes a structured, searchable career profile. 18K+ professionals, 100+ companies including Mercor, David AI, Rho and Warp.
Overall
Top 25%
Team read
Top 25%
Founder model
7.7x vs 1.9% base
ARR
$258,973 est
Visits
-
Sahil SethCEOBuilt an AI interviewer used by billion-dollar recruiting firms, led AI recruiting transformations at Bain. Northwestern summa cum laude.
Maki PotamianosCPOFounded an edtech marketplace in college; dual degree in applied math and finance.
Tony XavierCTOBuilt an open-source no-code editor; top 0.1% nationally in India in math and physics. UMich.
TechnicalBelow median
ShippingTop 25%
SellingTop 25%
Domain fitAbove median
Self-awareTop 25%
FocusAbove median
Taste readsThe widest kind of split, a16z engages while Vela passes. Founder model reads 7.7x.
Who they are up against
Mercor$135M+ raisedAI-scored talent marketplace, both partner and looming competitor.
Micro1venture-backedAI interviewer for technical hiring.
One endpoint with per-tool permissions across MCP, OpenAPI and GraphQL, collapsing 1,600+ tools into a ~1K-token surface. 2,800+ stars, 14K npm downloads a week, 3,000 local users.
Overall
Top 25%
Team read
Top 25%
Founder model
3.5x vs 1.9% base
ARR
-
Visits
-
Rhys SullivanCEOEx Vercel engineer, earlier Microsoft, Raven Software and Epic Games. Grew answeroverflow.com to 1.5M monthly users alone, bootstrapped.
TechnicalTop 25%
ShippingTop 10%
SellingBelow median
Domain fitAbove median
Self-awareTop 25%
FocusBelow median
Taste readsa16z engages on the adopted OSS artifact and solo shipping velocity. Vela's taste holds at in-between on solo risk in a lane the big harnesses could absorb.
Who they are up against
Zapier$1.4M raised, ~$5B valThe integration gravity well moving into agents.
Composio$31M raisedAgent tool access, the closest funded rival.
Arcade.dev$72M raisedAgent auth and tool-calling platform.
SF · team of 2 · partner Jared Friedman · $1.1M seed Jun 2026
Maps procurement, logistics, assembly and distribution, then executes them, from customs to carrier quotes. $15M+ in freight already handled for robotics and biochemical customers.
Overall
Top 25%
Team read
Above median
Founder model
2.6x vs 1.9% base
ARR
$227K
Visits
11.7K
Oguzhan KaracaCEOOxford law, legal at CMA CGM/Ceva (third-largest shipper), built Ivy Consultants to six-figure ARR on AI autopilot. Grew up in a farming family on the customer side of freight.
Mert TurnaCTOCTO of Hemut (YC P25, trucking ops), built ship-operations software running on 30+ vessels worth $200M+. Son of a ship captain, UCLA CS.
TechnicalBottom 25%
ShippingAbove median
SellingTop 25%
Domain fitAbove median
Self-awareTop 25%
FocusAbove median
Taste readsFreight actually handled plus a decade-long founder friendship, an AI-workforce wedge the model prices against labor.
Who they are up against
Flexport$2.7B raisedThe digital freight forwarding incumbent.
Altana$322M raisedSupply chain intelligence at scale.
Traditional freight forwardersn/aThe labor-heavy incumbents whose margin Derya's agents attack.
Predicts the wars, strikes and tariffs that break shipping one to three weeks early, tells freight teams which shipments are exposed and what to do. Correctly called the Hormuz ceasefire dates.
Overall
Top 25%
Team read
Top 25%
Founder model
3.8x vs 1.9% base
ARR
-
Visits
-
Alina ParkCEOFounding engineer at Inca Digital building a DARPA-funded AML platform deployed across DoD, CFTC and PayPal. CFTC Innovator of the Year, built SOTA text-geolocation at Yachay AI.
TechnicalTop 25%
ShippingAbove median
SellingAbove median
Domain fitTop 10%
Self-awareTop 25%
FocusAbove median
Taste readsa16z engages on the earned secret, eight years of escalation prediction now aimed at freight. Vela's taste holds at in-between pending revenue evidence.
Open-source knowledge graph engine with 108K+ GitHub stars and ~5M downloads; enterprise layer reviews every pull request and formally verifies code changes.
Overall
Top 25%
Team read
Above median
Founder model
0.9x vs 1.9% base
ARR
-
Visits
-
Safi ShamsiCEOBuilt Graphify to 100K+ GitHub stars and 4M+ PyPI downloads in about four months, first YC company to cross 100K mid-batch. Published researcher, MS Data Science Birmingham.
Deploy to secure enclaves in three commands and give customers and auditors cryptographic proof that production runs the exact reviewed code. In production, onboarding crypto, fintech, and private-AI customers.
Overall
Above median
Team read
Top 10%
Founder model
5.4x vs 1.9% base
ARR
-
Visits
-
Anton LivajaCEOCo-founded Distrust, the security consultancy behind systems holding $600B+ in assets. Before that OpenZeppelin research manager and founding-team security lead at Ledn.
Lance VickCTOSecurity lead at Turnkey, BitGo, Unit 410, and Pebble. Founded the Stagex hardened Linux distribution. Two decades in supply-chain security, HSMs, and key management.
Ksenia LeskoCOORan strategy and ops at Distrust, before that four years scaling customer education and AI adoption at Ada. Turns deep security engineering into something enterprises can buy.
TechnicalTop 10%
ShippingAbove median
SellingAbove median
Domain fitTop 10%
Self-awareTop 10%
FocusBelow median
Taste readsElite, verifiable domain pedigree, and the model hesitates on the consultancy-shaped history. The open question is whether Distrust's client trust converts into platform revenue rather than services revenue.
Echo decides how much compute each request needs, which open models work on it, and how answers combine. MATH-500 at 98.6% for $4.58 versus a frontier model's 99.8% at $12.64, evals published openly.
Overall
Above median
Team read
Top 25%
Founder model
5.3x vs 1.9% base
ARR
-
Visits
-
Adam RidaCEOEx ML interpretability researcher at Sorbonne/CNRS with published work, built applied ML across finance and insurance, grew DeepRecall to about 100K EUR ARR solo in under three months. Tracer's OSS crossed 1,000 GitHub stars.
TechnicalTop 25%
ShippingAbove median
SellingBelow median
Domain fitAbove median
Self-awareTop 10%
FocusBelow median
Taste readsA live product, published evals, 1,000+ stars, and a founder combining research depth with a prior revenue sprint. Solo status is priced in at a smaller check.
Who they are up against
OpenRouter$153M raisedThe model gateway Echo's single interface competes with.
Not Diamond$2.3M raisedModel routing intelligence, the closest research sibling.
Together AI$1.33B raisedOpen-source model serving at scale.
AI agents plus in-house lobbyists that monitor regulations, lobby policymakers, and win government contracts, owning the path from research to funded contract. Past a $500K revenue run-rate six weeks after quiet launch.
Overall
Above median
Team read
Above median
Founder model
4.8x vs 1.9% base
ARR
$500K
Visits
-
Parth BadhwarCEOFounding head of GTM at Clipbook (Mark Cuban backed), policy stints at the Biden White House drug policy office and Deloitte, campaign roles before that. GWU political science.
TechnicalBottom 25%
ShippingAbove median
SellingTop 10%
Domain fitAbove median
Self-awareTop 25%
FocusAbove median
Taste readsExceptional claimed velocity, but the model wants the $500K run-rate verified as recurring revenue and a named technical builder before committing.
Models each institution's full on-chain environment and evaluates every proposed transaction against live policy before execution. POCs with asset managers moving billions weekly, in evaluation at one of the world's largest banks.
Overall
Above median
Team read
Top 10%
Founder model
7.2x vs 1.9% base
ARR
-
Visits
-
Dan DanayCEOHead of blockchain security at Check Point, founded keyTango and Airmada (YC W16), MIT Sloan MBA, Israeli Navy missile-ship commander.
Isaac GrossmanCTOVP Engineering at Fhenix building an FHE L2, scaled Secret Network to a top-100 crypto asset at SCRT Labs, Matzov cryptography unit veteran, keyTango co-founder.
Liran KoganCPOLaunched blockchain security products at Check Point, led payments and FX products at Payoneer. NYU and INSEAD.
TechnicalTop 25%
ShippingAbove median
SellingTop 25%
Domain fitTop 10%
Self-awareTop 25%
FocusAbove median
Taste readsTwo hard gates fire, Tel Aviv headquarters with no visible US flip, and a crypto-native product the profile routes to a separate crypto fund. Team quality was not the issue.
Who they are up against
Fireblocks$1.04B raisedDigital asset custody incumbent at an $8B valuation.
Chainalysis$536.6M raisedOn-chain compliance and intelligence incumbent.
Hypernative~$58M raisedReal-time on-chain threat prevention, the closest direct rival.
Agents source and interview hundreds of experts in any language, combine it with public and fund data, and simulate the market, replacing $300-500K commercial diligence studies. Working with leading funds.
Overall
Above median
Team read
Above median
Founder model
3.9x vs 1.9% base
ARR
-
Visits
-
Tymek StaniszewskiCEOSix years in growth and private equity at Verdane, an $11B AUM fund. Lived the buyer's seat for the exact product he now sells.
Tien ChuCTOML and software across enterprise R&D, research labs and early startups. CS, math and philosophy at University of Warsaw.
TechnicalBelow median
ShippingBelow median
SellingTop 25%
Domain fitTop 25%
Self-awareBelow median
FocusAbove median
Taste readsa16z engages on the insider-plus-builder pairing from Verdane. Vela's taste holds at in-between on services-shaped delivery.
Who they are up against
AlphaSense~$1.4B raisedMarket intelligence incumbent that absorbed Tegus.
NewtonX$47M raisedExpert network for B2B research.
Strategy consultanciesn/aThe $500K-study status quo LATO undercuts.
SF · solo founder · team of 2 · partner Gustaf Alstromer · founded 2026
Private foundation models for banks and insurers predicting credit, fraud, churn and LTV inside the customer's cloud. One model already trained on 28B transactions.
Spin up teams of specialized agents that communicate, unblock and manage each other, so you are not babysitting a dozen chat windows. Third company for a pair that built Branch past $100M ARR together.
Overall
Above median
Team read
Top 25%
Founder model
6.4x vs 1.9% base
ARR
-
Visits
-
Mike MolinetCEOCo-founder, COO and President of Branch, zero past $100M ARR and a $4B valuation. Then co-founded Thena.ai and ran it three years. Verified end to end on LinkedIn.
Govind KavaturiCo-founderCo-founded Thena with Mike, early at Branch before that. The pair arrives intact.
TechnicalBelow median
ShippingTop 25%
SellingTop 10%
Domain fitAbove median
Self-awareTop 25%
FocusAbove median
Taste readsThe Branch record is verified and heavyweight, but the model wants the co-founder's technical role confirmed before treating this as a complete team.
Who they are up against
Glean$768M raisedEnterprise AI assistant incumbent expanding into agent teams.
Dust$21.5M raisedAI teammates wired into company tools.
Lindy$50M raisedAI employees for business workflows.
SF · team of 3 · partner Gustaf Alströmer · founded 2025
Connects fragmented test hardware, records every run with its context, and gives engineers an API to orchestrate the whole setup. Aerospace programs use it daily; an energy pilot targets $100K+ annual savings.
Overall
Above median
Team read
Top 25%
Founder model
1.0x vs 1.9% base
ARR
-
Visits
-
Darell ChuaCEOSystems performance architecture at Rivian, member of technical staff at DSO National Labs Singapore, UCSD MS. Met his co-founder testing liquid rocket engines in the Mojave.
Anchit KumarCTOUCSD CS, led AI and autonomy for Mars cave exploration drones, multimodal AI safety for robot models, 10+ Mojave test campaigns.
TechnicalAbove median
ShippingAbove median
SellingBelow median
Domain fitTop 25%
Self-awareTop 25%
FocusTop 25%
Taste readsa16z engages on lived founder-market fit and paying deployments. Vela's taste holds at in-between on the size of the wedge.
Who they are up against
Nominal$182.5M raisedTest and telemetry platform for hardware programs, the heavily funded direct rival.
Sift$67M raisedTelemetry stack for machines.
NI (Emerson)incumbentThe LabVIEW-era tooling Daqstra wraps.
Captures AI work traces, trains smaller models, and swaps them in only when they beat the frontier model, cutting bills by 80%.
Overall
Above median
Team read
Above median
Founder model
4.3x vs 1.9% base
ARR
$293,900 est
Visits
-
Luis ManriqueCEOFounding MTS and founding sales at Gumloop closing ~$2M year one. First PM on Instacart's Carrot AI, early Wildfire (acquired by Google), 5 years Google ads ML, SVP at VideoAmp.
Aamir PoonawallaCTO9 years at Instacart building ads serving, auction and experimentation infrastructure. 2nd-time YC founder. Georgia Tech MS CS.
TechnicalAbove median
ShippingAbove median
SellingTop 25%
Domain fitAbove median
Self-awareTop 25%
FocusBelow median
Taste readsRare double engage, both taste models lean in.
Who they are up against
OpenPipeacquired by CoreWeaveFine-tune-and-replace for LLM cost cutting, proven exit.
Tag an AI in Slack, hand it a task, get the result in the thread from whichever model is best that day. Live with 10 early-access teams roughly eight weeks after incorporation.
Overall
Above median
Team read
Top 25%
Founder model
6.9x vs 1.9% base
ARR
-
Visits
-
Tony KamCEOBerkeley EECS. Two stints shipping vehicle software at Tesla, Intel automation, CTO of a stealth energy-to-compute startup, then 2+ years co-running Lilac Labs (YC S24, voice AI for drive-thrus). Launched CubeSats in high school.
Shelden ShiCTOBerkeley CS and CogSci. Co-founder/CTO of Lilac Labs, a stealth startup before that, founding engineer at a fintech lending platform, internships at Flatiron Health, Roche, and CertiK.
Wilson NguyenChief EngBerkeley EECS, Head TA for CS162 Operating Systems. Founding engineer at Vestible, then founding engineer at Lilac Labs. The rare engineer who has seen two zero-to-ones from day one.
TechnicalAbove median
ShippingTop 10%
SellingAbove median
Domain fitAbove median
Self-awareTop 25%
FocusAbove median
Taste readsAn intact, fast team, but the taste model reads the lane as crowded and the harness as thin relative to its artifact bar. The strongest counterargument is the team's shipping velocity.
Who they are up against
Glean$768M raisedEnterprise AI assistant with company-wide context, the incumbent gravity well.
Lindy$50M raisedAI employees and assistants for business workflows.
Dust$21.5M raisedAI teammates wired into company tools, the closest architectural sibling.
SF · team of 2 · partner Tyler Bosmeny · founded 2025
Reads every conversation an agent has to find silent failures, then trains a specialist model on those traces. First customer model beat a frontier model with 22.9% higher task success at 94.5% lower cost.
Overall
Above median
Team read
Above median
Founder model
5.0x vs 1.9% base
ARR
$406K
Visits
-
Shubham PalriwalaCEOYoungest engineer on Cisco's analytics team, first hire at Formbricks, contributor to Bitcoin, OWASP and the Linux Foundation.
Parth AjmeraCTOIIT Madras CS, ranked 159 of over a million candidates. Led graphics engineering at Infurnia, built terabyte-scale pipelines at Microsoft. Childhood friends for eight years.
TechnicalAbove median
ShippingAbove median
SellingAbove median
Domain fitAbove median
Self-awareTop 25%
FocusAbove median
Taste readsBoth tastes land in-between. Real benchmark receipts and users at Google and Exa, but both models want the pivot from analytics to model training proven with recurring revenue.
Synchronized screen recordings, actions and annotations from real expert workflows in CAD, engineering and finance. 33K+ hours sold to frontier labs, 200K+ downloads on open datasets.
Overall
Above median
Team read
Above median
Founder model
1.8x vs 1.9% base
ARR
-
Visits
-
Dev MandalCEOPreviously at Sarvam, aerospace engineering at IIT Madras.
Harish AshokCTOFounded Zenith, a robotics company building AI-assisted hardware workflow tools.
TechnicalBottom 25%
ShippingAbove median
SellingTop 25%
Domain fitBelow median
Self-awareBelow median
FocusAbove median
Taste readsThe sharpest taste split in this cohort. a16z engages on real revenue and lab relationships, Vela's taste rejects the lab-data category the way it rejected Petrarch and Magma.
Who they are up against
Scale AI$15.9B raisedThe incumbent lab data pipeline.
Mercor$484M raisedExpert data marketplace.
Surge AIbootstrapped, ~$1B rev reportedProof the market pays and how crowded it is.
An on-call agent that drops read-only probes into running services and captures exact variable state at the moment of failure, no redeploys. Four-hour war rooms into ten-minute resolutions.
Overall
Above median
Team read
Above median
Founder model
1.5x vs 1.9% base
ARR
$259K
Visits
-
Shailendra SinghCEOFounded HyperProbe in 2022 (as HyperTest) and Transporter.city before that, a VC-backed logistics platform he wound down while returning investor capital. Two decades of operating scar tissue, verified on LinkedIn.
Karan RainaCTOCTO at HyperTest, senior architect at a SaaS company before, three years building production runtime SDKs.
TechnicalAbove median
ShippingAbove median
SellingBelow median
Domain fitAbove median
Self-awareTop 10%
FocusAbove median
Taste readsA working artifact with paying customers from the HyperTest era and founder-market fit earned over four years of runtime instrumentation.
Who they are up against
Sentry$217M raisedError monitoring incumbent every dev team already runs.
Lightrun~$110M raisedLive debugging and dynamic instrumentation, the closest direct competitor.
Agent swarms do the portal logins, report pulls, reconciliations and form-filling finance ops teams do by hand, with approval gates before anything submits.
Overall
Above median
Team read
Above median
Founder model
10.1x vs 1.9% base
ARR
$258,973 est
Visits
-
Jihyun KimCEOFounding team and Head of Product launching a $1B hedge fund, scaled Asia's largest wealth tech past $10B AUM. Ex-Goldman. Second-time founder. Cornell.
Varun ArumugamFounderFounding ML engineer at Bronco AI, Apple verification infrastructure. UT Austin ECE.
Cashel FitzgeraldCTOCornell MS CS, trains VLMs for computer use, SOTA aerial detection at Modern Intelligence.
Fleets of self-improving AI scientists running thousands of parallel ML experiments, matching SOTA ARC-AGI at up to 10,000x lower cost.
Overall
Above median
Team read
Above median
Founder model
1.9x vs 1.9% base
ARR
-
Visits
-
Joshua DillonFounder14 years Google/DeepMind/Luma staff research scientist. Created TensorFlow Probability, built Veo's first prototype, Gemini contributor, 11,600+ citations.
Dan KondratyukFounderFirst author of VideoPoet (ICML 2024 Best Paper), led Luma's World Models team.
TechnicalTop 10%
ShippingAbove median
SellingBottom 25%
Domain fitTop 10%
Self-awareBelow median
FocusBelow median
Taste readsRare double engage, both taste models lean in.
Who they are up against
Sakana AI$200M+ raisedThe AI Scientist agenda at lab scale.
FutureHousenonprofit, fundedAutonomous research agents for science.
Watches fleets at the kernel level through syscalls and eBPF traces, catches privilege escalation and memory corruption other layers miss, and lands the fix fleet-wide in minutes. Already handles real CVEs.
Overall
Above median
Team read
Below median
Founder model
9.1x vs 1.9% base
ARR
-
Visits
-
Fnu PrinceCEOLeft Caltech at 18. Professional red-team and blue-team work since 16, USACO Platinum, USAPhO honorable mention, built his own OS before code assistants existed.
TechnicalTop 10%
ShippingBottom 25%
SellingBottom 25%
Domain fitAbove median
Self-awareBottom 25%
FocusBelow median
Taste readsA textbook young-prodigy speedrun profile. Kernel-level receipts, an OS built from scratch and USACO Platinum substitute for the missing co-founder.
Who they are up against
CrowdStrikepublicEndpoint security incumbent.
Wiz$2B raised, acq. GoogleCloud security platform that set the category's pace.
Oligo Security$78M raisedRuntime application security via eBPF, the closest technical sibling.
One API and MCP giving agents cited answers about any US location across 300+ fields of land, power, people and policy, with on-demand indexing for anything missing. First customers source data-center land 100x faster.
Overall
Above median
Team read
Below median
Founder model
5.1x vs 1.9% base
ARR
$194K
Visits
-
Ansh ChokshiCEOCo-founded Poker Pit at 21 and sold it after reaching $250K ARR bootstrapped in four months, verified on LinkedIn. Then three years at Seismic as data scientist and engineer, a stint at Eraser before founding.
Shashwat KapoorCo-founderCo-founder, joined for the S26 build.
TechnicalBelow median
ShippingAbove median
SellingAbove median
Domain fitBelow median
Self-awareBelow median
FocusAbove median
Taste readsThe Poker Pit exit is a real velocity receipt, but solo status without geospatial history leaves the model wanting a technical co-founder or verified customers first.
Who they are up against
Cherre$105M raised, acquiredReal estate data integration platform.
RegridbootstrappedNationwide parcel data incumbent.
County GIS and PDFsn/aThe real competitor is the fragmented status quo Mireye indexes.
SF · team of 3 · partner Tyler Bosmeny · founded 2025
Every agent gets its own email, phone, iMessage, internet address and credential vault through one API, so agents can act in the world with a persistent identity.
Overall
Above median
Team read
Below median
Founder model
1.6x vs 1.9% base
ARR
-
Visits
-
Ray LiaoCEOMIT CS PhD. Previously co-founder and CTO of Empallo, an AI virtual clinic for cardiovascular care.
Alex WilcoxCTONortheastern MS CS with ML engineering and computer networking research.
Dima VremenkoCAIOHarvard Medical School research associate, ML on gigapixel pathology images, six publications.
TechnicalAbove median
ShippingAbove median
SellingBelow median
Domain fitBelow median
Self-awareBelow median
FocusAbove median
Taste readsA rare double engage. Both tastes read agent identity as core control-plane infrastructure and credit the MIT PhD plus repeat-founder pairing.
Who they are up against
TwiliopublicThe communications API incumbent built for humans.
Composio$31M raisedAgent tool access and auth.
Browserbase$67.5M raisedAgent infrastructure for acting on the web.
Open-source, local-first recording of screen and audio that becomes searchable memory and agent context. 20K+ GitHub stars.
Overall
Above median
Team read
Below median
Founder model
2.0x vs 1.9% base
ARR
-
Visits
-
Louis BeaumontCEOPreviously French Intelligence (acquired). OpenAI RL OSS contributor a decade ago, video game bot business at 14. Turned down early OpenAI, Cursor, Mercor.
TechnicalAbove median
ShippingTop 10%
SellingBelow median
Domain fitBelow median
Self-awareBottom 25%
FocusBottom 25%
Taste readsa16z engages, Vela stays in-between.
Who they are up against
Limitless (Rewind)$33M+ raisedContinuous personal recording with AI recall.
Highlight AIventure-backedOn-device context capture for assistants.
AI employees interview each role on the team, map workflows, connect to the existing stack and do the work 24/7 with humans approving what matters. Live with a Fortune Global 100 customer.
Overall
Above median
Team read
Top 25%
Founder model
2.4x vs 1.9% base
ARR
-
Visits
-
Jibril MoinuddinCEOFounding engineer at Buckeye AI and TrueFoundry, founded STEMplicity Education, Michigan dropout, Non-Trivial Fellow.
Christ XuCTOYoungest Tencent senior SWE, then OpenAI engineer. Two-time IOI camp, IMO camp, published algorithms researcher.
TechnicalTop 25%
ShippingAbove median
SellingAbove median
Domain fitAbove median
Self-awareBelow median
FocusAbove median
Taste readsa16z holds at in-between and Vela's taste rejects, both wanting the Fortune Global 100 deployment's revenue quality verified against batchmates in the same lane.
Who they are up against
Billow AI Labs (same batch)YC S26Batchmate attacking the same accounting lane as a firm.
Voice receptionist, a custom OBD-II scanner feeding a diagnostic agent, instant estimates from real parts pricing, and service reminders, one system replacing a web of disjointed software. Live in 40+ shops across 18 states.
Overall
Above median
Team read
Above median
Founder model
0.9x vs 1.9% base
ARR
-
Visits
-
Michael MuzzinCEOElectrical engineering at NVIDIA, authored the factory failure disposition guide governing re-qualification of $300M in HGX and DGX hardware. UCLA BSEE.
Junaid PopalzaiCTOEngineering at Netflix, Notion, Whatnot, Tesla and Amazon. Berkeley EECS.
TechnicalAbove median
ShippingTop 25%
SellingTop 25%
Domain fitBelow median
Self-awareBelow median
FocusAbove median
Taste readsStrong vertical fit and live shops, but the model could not verify the CTO's commitment or the activation quality of the 40 deployments from public data alone.
Describe a goal and Nex builds the workflow, writing code for the deterministic parts and saving model judgment for the hard parts. Backed by HubSpot founder Dharmesh Shah.
Overall
Above median
Team read
Top 10%
Founder model
1.7x vs 1.9% base
ARR
$574K
Visits
-
Najmuzzaman MohammadCEOSenior PM at HubSpot and Clipboard Health, founding hire at VentureBasecamp (acquired by Wadhwani Foundation within 2 years), profitable e-commerce business in college.
Francisco DiasCTO4th engineer on HubSpot CRM pre-IPO, first product engineer on HubSpot's AI infrastructure team, earlier shipped BostonGlobe.com. 22 marathons and a sub-11 Ironman.
TechnicalAbove median
ShippingAbove median
SellingTop 10%
Domain fitTop 25%
Self-awareTop 10%
FocusTop 25%
Taste readsThe model could not verify the co-founder pairing from public evidence and reads the GTM-agent lane as crowded. The HubSpot pedigree and Dharmesh Shah backing are real counterweights.
Who they are up against
Clay$202M raisedThe GTM data and workflow platform every RevOps team already tried.
11x$76M raisedAI SDR agents, the adjacent lane with heavy funding.
Artisan$46.1M raisedAI employees for outbound sales.
Pulls sessions from Claude Code, Codex and Cursor plus Slack, PRs, tickets and calls, then injects prior context into every new agent task.
Overall
Above median
Team read
Below median
Founder model
8.4x vs 1.9% base
ARR
$193,733 est
Visits
-
Nikos DritsakosCEOBootstrapped SalesBop to a $500K acquisition in 18 months, VP Technology at FliteHouse at 22, ex Head of Special Projects at Composio.
One API key routes each call to the best STT, LLM and TTS for the language and use case, re-benchmarked publicly the day new models ship.
Overall
Above median
Team read
Top 25%
Founder model
2.1x vs 1.9% base
ARR
-
Visits
-
Beknazar AbdikamalovCEOCo-founder/CTO of Hupo ($14M raised, DST and Meta backed), 4 years shipping production voice for Morgan Stanley, Prudential, HSBC, Grab in five languages. Ex-Amazon.
TechnicalAbove median
ShippingAbove median
SellingBelow median
Domain fitTop 25%
Self-awareTop 10%
FocusTop 25%
Taste readsa16z engages, Vela stays in-between.
Who they are up against
Vapi$130M raisedVoice agent orchestration platform with scale.
Sources codebases, project files and operational data from companies in bankruptcy and pivots, de-identifies it, certifies legal clearance, and resells to labs. Already transacted construction plans, AV recordings, and codebases.
Overall
Below median
Team read
Below median
Founder model
14.0x vs 1.9% base
ARR
$259K
Visits
-
Ian LeeCEOHarvard applied math dropout. Fellow at Meridian AI building training data infrastructure, technical advisor at Scale AI on quality filtering for reasoning models.
Samuel HahnCOOHarvard economics dropout. Led GTM at Magier AI (Techstars 24) on PII redaction, venture analyst before that.
Sudhish SwainCTOHarvard bio dropout. Head of software engineering at Clean Sweep Group, computer vision deployed in California hospitals, cancer dataset research at Brigham.
TechnicalBottom 25%
ShippingBottom 25%
SellingAbove median
Domain fitAbove median
Self-awareBelow median
FocusAbove median
Taste readsTraining-data supply for labs is a zero-deal category in the profile without frontier standing. The founder model loves this team at 14x while the taste model rejects the category, the sharpest disagreement in the cohort.
GitNexus indexes a repository into a knowledge graph and serves agents exactly the context a task needs. On DeepSWE the same agent goes from 37% to 68.4% pass rate at half the cost.
Overall
Below median
Team read
Below median
Founder model
1.4x vs 1.9% base
ARR
-
Visits
-
Subham KunduCEOCTO at Cignara (YC P26), principal AI engineer at HTCD, founding engineer at Caravel Labs, RL PhD ongoing.
Abhigyan PatwariCTOSerial open source contributor, roughly five years working on knowledge graphs.
TechnicalAbove median
ShippingAbove median
SellingBelow median
Domain fitAbove median
Self-awareBottom 25%
FocusBelow median
Taste readsa16z engages on claimed OSS traction and benchmark receipts. Vela's taste rejects pending independent verification of the star and download claims.
Who they are up against
Cursor (Anysphere)$3.4B raisedThe coding-agent gravity well building its own context engines.
One place to deploy, govern and maintain every AI agent in a company. Scoped credentials per agent, approval gates on sensitive actions, full audit trails, and retirement for agents nobody uses.
Overall
Below median
Team read
Above median
Founder model
6.0x vs 1.9% base
ARR
-
Visits
-
Aman RajCEOBuilt the AI compliance platform at Barclays used globally, founded a fintech that scaled to 52K families and $1.5M, led product at a consumer AI app with 100K+ MAU (Google Play Best App 2024). IIT Kharagpur.
Sarthak AggarwalCTOAI systems at NVIDIA deployed at OpenAI and Meta, enterprise AI at Ema, Google Code-In global winner, open source since age 12. BITS Pilani.
TechnicalAbove median
ShippingBelow median
SellingAbove median
Domain fitAbove median
Self-awareBelow median
FocusAbove median
Taste readsThe control-plane thesis fits, but the model flagged unverified incorporation details and profile inconsistencies it wants resolved.
Who they are up against
Zenity$59.5M raisedAgent and copilot security governance, the funded direct rival.
Microsoft agent governanceincumbentShipping governance natively for its own agent ecosystem.
Not voice swap but scene understanding, who is speaking, where each sound originates, what survives the cut. Published benchmarks beat the incumbent on background-sound error by 270% and voice similarity by 27.8%.
Overall
Below median
Team read
Above median
Founder model
1.6x vs 1.9% base
ARR
-
Visits
-
An Zhu LiuCEOStanford dropout who sold a voice AI product for six figures and turned down an eight-figure offer to lead model research at a unicorn.
Mingi KwonCo-founderAI PhD from Yonsei, text-to-video at Adobe Research, ran YouTube channels with 400K+ subscribers.
Xu ZhengCo-founderAI PhD from HKUST, 30+ top-venue publications, 2K+ citations. The team counts 10,000+ citations and 100+ papers.
TechnicalTop 10%
ShippingAbove median
SellingBelow median
Domain fitAbove median
Self-awareBelow median
FocusBelow median
Taste readsBoth tastes hold at in-between. Extraordinary research depth against a Toronto base and an ElevenLabs-shaped giant, with commercial proof still to come.
Who they are up against
ElevenLabs$942M raisedThe AI audio giant whose dubbing Familiar benchmarks against.
HeyGen$69M raisedAI video translation and avatars.
Deepdub~$20M raisedAI dubbing for studios, the closest direct rival.
Describe a task in plain English; a custom vision model watches the feed and operates a computer when it sees stalled production, safety violations or waiting customers.
Self-improving agents that learn a distributor's business and run inventory planning, customer service, and order processing in a $60T manual industry.
Overall
Below median
Team read
Top 25%
Founder model
1.4x vs 1.9% base
ARR
-
Visits
-
Alex TungCEOFirst non-engineer hire at Index-backed Ankar, scaling it to 7-figure ARR in under a year. Ex-BCG deploying AI at F500 distributors. Cambridge Economics.
Leon YaoCTOResearch Engineer at InstaDeep running 7-figure supply chain AI deployments for F500 enterprises. UCL ML, published at IROS as an undergrad.
TechnicalBelow median
ShippingBelow median
SellingTop 25%
Domain fitAbove median
Self-awareTop 10%
FocusTop 25%
Taste readsa16z holds at in-between, Vela stays in-between.
Who they are up against
Ferry (distribution AI)venture-backedAI order entry for distributors, the narrow wedge version of the same buyer.
Pantheon / legacy ERP add-onsincumbentsEpicor and Infor own the distributor back office Whitespace wants to replace.
Factory-direct China-to-US imports at one locked landed price, agents mining public customs records, own inspectors on factory floors.
Overall
Below median
Team read
Above median
Founder model
2.2x vs 1.9% base
ARR
-
Visits
-
Benjamin MartindaleCEOFounded Prelude OS, 31 paying Chinese factory customers cold-acquired in 39 days in Guangzhou, cash-flow positive. Advised Sany and Sinoboom. Fluent Mandarin. Featured on Mango TV.
Minghao TanCTOParses raw customs filings, separates real factories from shells and forwarders.
Model Processing Units burn LLM weights directly into the chip, targeting 20,000+ tokens per second at 63x higher intelligence per watt. An RL environment co-designs the chip architecture for each model.
Overall
Below median
Team read
Below median
Founder model
7.4x vs 1.9% base
ARR
-
Visits
-
Niki KotechaCEOAI PhD at Imperial College and I-X Scholar, Cambridge MEng, Alan Turing Institute global fellowship, 160+ citations.
Thomas LanningCTOOxford masters in mathematical and theoretical physics, top of class at Edinburgh, rediscovered the Higgs boson using ML on CERN data.
TechnicalTop 10%
ShippingBottom 25%
SellingBottom 25%
Domain fitAbove median
Self-awareBelow median
FocusAbove median
Taste readsSilicon is a zero-deal category in the profile without frontier-lab standing, and the London base fires the US-headquarters gate.
Who they are up against
Groq$2.4B raisedDeterministic inference chips at scale.
Etched$925M raisedTransformer-specialized ASICs, the closest thesis rival.
Sells the same service as a traditional firm while agents do the work on a proprietary general ledger and a certified accountant signs off. Helsinki-built, aimed at startups and SMBs in Europe and the US.
Overall
Below median
Team read
Below median
Founder model
7.7x vs 1.9% base
ARR
-
Visits
-
Sami LaakkonenCEOBuilt petabyte-scale data infrastructure for Google and Bloomberg clients at Token Terminal. Two-time Nordplus physics olympiad, ex fighter pilot training, CS dropout.
Riki KarjalainenCo-founderToken Terminal data infrastructure, Aalto applied mathematics.
Veikko PereCo-founderBCG and private equity at Intera Partners, finance and law degrees.
Pietari VirkkunenCo-founderEngineering and sales across tech, VC and PE, Aalto industrial engineering.
TechnicalAbove median
ShippingBelow median
SellingAbove median
Domain fitAbove median
Self-awareBelow median
FocusAbove median
Taste readsBoth tastes are negative, a16z on the Helsinki headquarters gate and Vela on a crowded agent-accounting lane already carrying two batchmates.
An open-source customization layer that reads a product's theme, components and API, then lets end users describe features in plain English and get them live inside the app, within existing guardrails.
Overall
Below median
Team read
Below median
Founder model
2.2x vs 1.9% base
ARR
-
Visits
-
Nour ZahzahCEOBerkeley MechE, previously founded an in-store AI shopping assistant startup with his co-founder. Seven years of professional academy soccer at SJ Earthquakes.
Yousef HelalCTOGraduated high school at 15, Berkeley EECS top of class at 19, Stanford MS. Apple FPGA infrastructure and AMD chip design. Kindergarten friends.
TechnicalAbove median
ShippingBottom 25%
SellingBelow median
Domain fitBelow median
Self-awareBelow median
FocusBelow median
Taste readsa16z engages on the young-prodigy pairing with a prior company together. Vela's taste rejects the horizontal wedge before adoption evidence.
Who they are up against
Retool$141M raisedInternal tools incumbent expanding into AI apps.
Appsmith$51.7M raisedOpen-source app builder.
In-product AI chatbotsn/aThe bolted-on status quo Vendo replaces.
Surfaces opportunities before the RFP drops, backtests win probability against millions of award outcomes, and drafts proposals on a company brain. Supports contractors with 100+ prime awards and $250M+ obligated value.
Overall
Below median
Team read
Below median
Founder model
0.8x vs 1.9% base
ARR
-
Visits
-
Ananth SankaralingamCEOUMD CS with NeurIPS and ICLR accepted LLM-compression research, engineering at LinkedIn and Amazon. Young profile with real research signal, verified on LinkedIn.
Akshay MistryCo-founderMet Ananth in a high school English class. Infrastructure at Google and Plaid per the launch post.
TechnicalBelow median
ShippingBelow median
SellingAbove median
Domain fitBelow median
Self-awareBelow median
FocusBelow median
Taste readsA desk-work vertical nobody on the team has lived, per the model. It wants a GovCon insider paired with the builders and reads the logo wall as batchmate pilots.
Who they are up against
GovDash$40M raisedAI for government contracting proposals, the funded direct rival.
DeltekPE-ownedThe GovCon software incumbent.
FiscalNotepublic, $738M raisedPolicy and procurement intelligence platform.
Ties AI spend to delivery per engineer, then cuts ~30% of it via routing, context optimization and prompt compression, without reading prompts or code.
Overall
Below median
Team read
Bottom 25%
Founder model
4.5x vs 1.9% base
ARR
-
Visits
-
Ashank ShahCEOCaltech CS, generative and graph DL research at Dartmouth Health.
Ahmet DemirbasCTOCaltech CS, 5x published in ML, NLP and agents at IEEE, ACM, BMC.
Imagination models that learn from internet video. Photon-1, a 106B MoE trained on 18 years of screen recordings, beats a production LLM on computer use with 30x less compute.
Overall
Below median
Team read
Above median
Founder model
4.5x vs 1.9% base
ARR
-
Visits
-
Jonathan LiCEOBuilding foundation models at Induction Labs.
David LiFounderBuilding foundation models at Induction Labs.
TechnicalTop 10%
ShippingAbove median
SellingBottom 25%
Domain fitAbove median
Self-awareBottom 25%
FocusAbove median
Taste readsa16z passes, Vela stays in-between.
Who they are up against
World Labs$230M raisedWorld models from video at frontier scale.
Moonvalley / Odysseyventure-backedVideo-native model labs racing the same thesis.
Say what part you need and it arrives, sourced, paid, customs-cleared, invoice-matched and logged into inventory. Asked to source a lifetime supply of an obsolete RF component, industrial robots, and actual Cessna aircraft in its first two weeks.
Overall
Below median
Team read
Above median
Founder model
1.0x vs 1.9% base
ARR
-
Visits
-
Rishi LaddhaCEOFounding team running operations, supply chain and government relations at LAT Aerospace (Deepinder Goyal's aerospace startup). Built AI trade assistants at INDmoney.
Tushar GoyalCo-founderBuilt the autonomous inventory system behind 400+ D2C brands at WareIQ (YC S20).
TechnicalBottom 25%
ShippingAbove median
SellingAbove median
Domain fitAbove median
Self-awareBelow median
FocusAbove median
Taste readsBoth tastes pass, a16z on the India headquarters gate, Vela on the services-heavy sourcing model, while both credit the lived procurement scar tissue.
Who they are up against
Cofactr$31.3M raisedSupply chain and logistics for electronics parts.
SDK-connected agent traces become autonomously trained private models, routers and harnesses. Live with 10 companies at up to 10x lower inference cost.
Overall
Below median
Team read
Below median
Founder model
5.2x vs 1.9% base
ARR
-
Visits
-
Zachary YuCEOSolo founder whose bio reads 'i make models.'
TechnicalAbove median
ShippingAbove median
SellingBelow median
Domain fitBelow median
Self-awareBottom 25%
FocusBelow median
Taste readsa16z passes, Vela passes.
Who they are up against
Understudy Labssame batchThe identical thesis two desks over.
OpenPipeacquired by CoreWeaveProof the trace-to-model wedge exits.
Routes every AI request bottom-up. Device first at zero API cost, efficient cloud second, frontier model last. A Rust engine they report decodes up to 60% faster than llama.cpp on Apple Silicon.
Overall
Bottom 25%
Team read
Bottom 25%
Founder model
4.3x vs 1.9% base
ARR
$294K
Visits
1.7K
Michael JeffordsCo-founderUT Austin, on a pre-med track until this year. Five years of stem-cell research at UT Health San Antonio, an ML pose-estimation pipeline for ALS and Parkinson's detection, EMT and ICU shifts, co-founded the HeartCheck Foundation which screened 10,000+ people.
Charles MuehlbergerCTOPrinceton engineering. ML research on multi-modal edge inference and physical neural networks, an ML engineering stint at the Department of War, quant research internship, and the world-record fastest electric speedboat team.
TechnicalAbove median
ShippingBelow median
SellingBelow median
Domain fitBottom 25%
Self-awareBottom 25%
FocusBottom 25%
Taste readsRaw talent is evident, but the taste model wants a shipped artifact at venture depth in a lane OpenRouter and the local runtimes already occupy. The strongest counterargument is the Rust engine benchmark, if it holds under scrutiny.
Who they are up against
OpenRouter$153M raisedThe cloud model-routing gateway, the default alternative.
Ollamaa16z-backedThe de facto local inference runtime Conifer must outperform.
Frontier lab defaultsn/aLabs route everything to their own cloud, the behavior Conifer monetizes against.
Plugs into CAD, ECAD, GitHub, and Jira, builds a live dependency graph of the robot, and when a part changes its agents trace the impact, run the tests, and propose fixes on branches. Onboarding via waitlist.
Overall
Bottom 25%
Team read
Bottom 25%
Founder model
3.4x vs 1.9% base
ARR
-
Visits
0.4K
Anish MariathasanCEOImperial College MEng in biomedical engineering, dropped an Oxford AI PhD. Co-founded Regulos (FDA compliance AI) months before Osseus, ML research at a Harvard-HKU cancer imaging lab.
Ayush PatelCTOImperial College MEng in computer science. Two Apple internships building agentic QA systems, one presented to Apple VPs with a recommendation for org-wide adoption.
TechnicalBelow median
ShippingBottom 25%
SellingBelow median
Domain fitBottom 25%
Self-awareBelow median
FocusBottom 25%
Taste readsPlatform breadth at waitlist stage, with no robotics operating history on the team yet. The model wants an artifact robots teams already depend on. The counterargument is the Apple-honed agent tooling DNA.
Every AI worker gets a job, a budget, guardrails and a measurable outcome, closing the loop from allocation to attribution so companies can spend tokens where they make money.
Overall
Bottom 25%
Team read
Below median
Founder model
1.7x vs 1.9% base
ARR
-
Visits
-
Hamze Al-ZamkanCEOScaled TUM.ai into Europe's largest student-led AI initiative, deployed AI at appliedAI, growth at Voggt and Terra.
Amgad Al-ZamkanCPOEx Amazon, CS at Berkeley and TUM, research on recommendation optimization.
Muhtasham OblokulovCTOEarly DeepSeek contributor, co-author of the StarCoder series with 3,000+ citations, ICML published, ex senior MLE at Munich Re where his models drove $1B in revenue.
TechnicalTop 25%
ShippingBottom 25%
SellingBelow median
Domain fitAbove median
Self-awareBelow median
FocusBelow median
Taste readsa16z holds at in-between and Vela's taste rejects. A rare CTO in a crowded horizontal lane, with the attribution thesis still unproven by revenue.
Who they are up against
Sierra$1.59B raisedOutcome-priced agents at scale, the thesis validator.
Lindy$50M raisedAI employees for business workflows.
Zapier$1.4M raised, ~$5B valWhere workflow automation budgets sit today.
Companies share agent traces read-only, Magma anonymizes them into training environments, licenses them to AI labs, and the company gets paid every time. Launched this week.
Overall
Bottom 25%
Team read
Bottom 25%
Founder model
9.6x vs 1.9% base
ARR
-
Visits
-
Ramtin ShahzanianCEOBuilt the first Persian OCR model and legal LLM (Roya), transformer architecture research, applied AI for industrial filtration. Three stealth attempts in the twelve months before Magma, described on his own profile as failing his way through.
TechnicalBottom 25%
ShippingBottom 25%
SellingBottom 25%
Domain fitBottom 25%
Self-awareTop 10%
FocusBottom 25%
Taste readsA two-sided data marketplace needs supply and demand it does not yet have, run by a solo founder with three short attempts in a year. The candor is a genuine positive signal the model does not price.
Who they are up against
Mercor$484M raisedThe scaled marketplace for human training data and evaluations.
Scale AI$15.9B raisedOwns the lab relationships Magma needs on the demand side.
Surge AIbootstrapped, ~$1B rev reportedProof the lab-data market pays, and how crowded it already is.
One MCP install gives agents research connectors the built-in web search misses, professional platforms, marketplaces, financials and live travel fares, without handing over a logged-in browser.
Overall
Bottom 25%
Team read
Below median
Founder model
1.0x vs 1.9% base
ARR
-
Visits
-
Aditya AsgaonkarCEOHelped build Ethereum's proof-of-stake as a researcher and open-source contributor at the Ethereum Foundation.
TechnicalTop 25%
ShippingBottom 25%
SellingBottom 25%
Domain fitBelow median
Self-awareBelow median
FocusBelow median
Taste readsa16z passes on a thin connector wedge with platform risk. Vela's taste holds at in-between out of respect for the protocol-research pedigree.
Who they are up against
Exa$357M raisedSearch API for AI agents.
Browserbase$67.5M raisedWeb access infrastructure for agents.
Native connectorsn/aOpenAI and Anthropic shipping their own connector directories.
An agent with its own computer and a self-updating company wiki compiled from Gmail, Calendar, GitHub, docs and meeting notes, shared across teammates.
Overall
Bottom 25%
Team read
Bottom 25%
Founder model
1.2x vs 1.9% base
ARR
-
Visits
-
Rohan SharmaCEOML research at Vector Institute, CMU and MIT. MEng Computational Science Harvard, BTech IIT Delhi.
Kushagra ChitkaraCOOMasters in Data Science Harvard '25, IIT Kharagpur '23.
Divit ShethCTOSenior Software Engineer at Arcesium. BITS Pilani.
TechnicalAbove median
ShippingAbove median
SellingBelow median
Domain fitBottom 25%
Self-awareBottom 25%
FocusBottom 25%
Taste readsa16z passes, Vela stays in-between.
Who they are up against
Almanac competes with every horizontal agentcrowdedPersonal-agent-with-memory is the most contested category in the batch.
Lindy$50M raisedPersonal AI employee with account access, the scaled version of this pitch.
Connects Pylon, Granola, Posthog, Hubspot, Slack and Notion into one living customer model, then answers churn and roadmap questions with cited evidence.
An AI forward-deployed engineer that learns each customized ERP and deploys agents into it, turning the $50B implementation-services market into software.
Overall
Bottom 25%
Team read
Below median
Founder model
3.5x vs 1.9% base
ARR
$258,973 est
Visits
-
Matthew ChowCEOSafety-critical software for Tesla Cybertruck and Zipline drones. Won Stanford's largest hackathon. Sales in Japan at Schneider Electric. UBC.
Victor VannaraCTOBuilt Amazon Payments' first AI support agent. Discord bots with 400K users at 13. UBC CS.
TechnicalBelow median
ShippingBelow median
SellingBelow median
Domain fitBelow median
Self-awareBelow median
FocusBelow median
Taste readsa16z passes, Vela passes.
Who they are up against
Assemblesame batchS26 rival for agents inside enterprise systems.
Accenture / DeloitteincumbentsThe services giants whose margin is the target.
Science-experimentation benchmarks that mirror real discovery, sold to frontier labs as RL environments. Pre-launch, the quietest company in this cohort.
Overall
Bottom 25%
Team read
Bottom 25%
Founder model
2.6x vs 1.9% base
ARR
-
Visits
-
Dillon MehtaCEOFounder and CEO. Thin public record so far, the launch post has not landed yet.
Neel GaddeCOOPreviously co-founder and CRO at Snap Poker.
Taste readsA plausible young-builder match stalled by unknowns, whether the product is evaluation or model-training infrastructure, and two unnamed co-founders. The model also noticed the company's LinkedIn description points to a pivot toward measuring employee AI usage.
Who they are up against
Prime Intellect$150.5M raisedOpen RL environments and decentralized training infrastructure.
Mercor$484M raisedExpert data and evals marketplace serving the same lab buyers.
Scale AI$15.9B raisedThe default data and RL environment vendor for frontier labs.
Every founder in the cohort maps to one or more of nine archetypes. Six are classics of seed investing and three earned their place from this batch's data. Click any card to filter the whole page by that persona.
05
How this was built, and why we share it
The pipeline starts at the YC S26 directory and covers every company listed under B2B in the YC S26 directory, the complete slice rather than a curated one. Founder histories come from LinkedIn and were checked against each company's YC profile. Model outputs come from Vela OS. The founder prediction model scores each founding team from their LinkedIn histories and the company idea, expressed as a multiple of the 1.9% baseline outlier rate. Two taste models score each lead founder, Vela Taste is the conviction profile of Vela, a tier-1 quant VC, and the a16z taste is built from the public track record of a16z, a tier-1 VC. Both verdicts are reported unedited, and where they disagree the split itself is the insight. ARR estimates carry wide confidence intervals at this stage and should be read as order-of-magnitude.
Contributing backVela is a tier-1 quant VC, and quant models get better when the ecosystem argues with them, so this page is public by design. The full B2B batch is now covered. If a correction or a new disclosure should change a row, write to yigit@vela.partners.
