AfterQuery

AfterQuery is a Y Combinator-backed applied research lab that sells expert-built training data, reinforcement-learning environments and evaluations to frontier AI labs and large technology companies. It pays practicing engineers, doctors, lawyers and finance professionals to encode how they reason, so model builders can target specific capability gaps with professional-grade data.

AfterQuery is a San Francisco applied research lab, founded in 2025 and part of Y Combinator's Winter 2025 batch, that sells expert-generated training data, reinforcement-learning environments and evaluations to frontier AI labs and large technology companies. Its model is to recruit practicing professionals in engineering, medicine, law and finance to encode how they reason: supervised fine-tuning sets of prompt-response pairs with chain-of-thought, rubric- and verifier-based RL data, RLHF preference comparisons, tool-calling environments built on real APIs and MCP servers, and computer-use and browser-use environments with human demonstrations. Delivery is either off-the-shelf datasets or custom evals and training sets scoped to a lab's capability gap. The expert network was described as nearly 100,000 professionals when it raised a $30 million Series A at a $300 million valuation in April 2026, led by Altos Ventures with Y Combinator, The Raine Group and BoxGroup; its products page now claims more than 300,000 verified practitioners. It reported more than $100 million in annualized revenue at that raise, and in September 2026 TechCrunch reported a valuation of $3.2 billion, which YC partner Gustaf Alströmer called the fastest launch-to-unicorn in YC history. The clearest public proof point is NVIDIA, which used AfterQuery's Office Agent Training Dataset to warm up Nemotron 3 Ultra for the GDPval benchmark; AfterQuery is the only data partner named in that model's technical report. TechCrunch also names Legora and Motif Technologies as customers. AfterQuery publishes its own benchmarks, including IDE-Bench, Market-Bench, App-Bench, FinanceArena and the VADER security benchmark, and competes with Scale AI, Mercor and Surge AI for lab data budgets.

Category

Data & Analytics

Pricing

Contact for pricing

Target Market

CTOs, AI Research Labs, ML Engineering Leaders, Data Scientists

Deployment

API-based

Founded

2025

Headquarters

San Francisco, United States

Team Size

11-50

Customers

Claims every US-based frontier AI lab; named customers include NVIDIA, Legora and Motif Technologies

• ✓
  Supervised fine-tuning datasets

  Expert-written prompt-response pairs with chain-of-thought demonstrations that build foundational skills before a model enters reinforcement learning.

• ✓
  Rubric and verifier-based RL

  Expert-authored grading rubrics paired with automated verifiers score reasoning, code and instruction-following outputs the way a seasoned practitioner would.

• ✓
  Tool-calling RL environments

  Custom environments built on real APIs, developer tools and MCP servers teach agents to call, chain and recover from tool errors.

• ✓
  Computer-use and browser-use environments

  Realistic desktop and browser interfaces paired with human expert demonstrations train agents to complete multi-step professional workflows.

• ✓
  Off-the-shelf datasets

  Pre-built packages such as the Office Agent Training Dataset, built from file-grounded professional tasks producing spreadsheets, slide decks and documents.

• ✓
  Public benchmarks and leaderboards

  Published leaderboards including IDE-Bench, Market-Bench, App-Bench and FinanceArena show buyers how AfterQuery measures frontier models before engaging.

• •
  Agentic office-work training

  A frontier lab warms up a large model on file-grounded professional deliverable tasks before RL, as NVIDIA did to lift Nemotron 3 Ultra's GDPval score from 35.3 to 46.7.

• •
  Domain-expert reasoning data

  A model team commissions chain-of-thought and preference data from practicing doctors, lawyers or finance professionals to close a specialist-domain capability gap.

• •
  Coding-agent RL environments

  An AI coding team builds tool-calling environments with automated verifiers so its agent learns to chain tools and recover from errors.

• •
  Custom model evaluation

  A lab or enterprise commissions custom evals scored against expert rubrics to measure a model on its own target professional tasks rather than only public benchmarks.

No list pricing is published on afterquery.com; every engagement starts through a 'Get data' sales contact. Sacra reports B2B contracts priced on domain difficulty, expert involvement, environment complexity and dataset volume rather than per seat, with multi-million-dollar annual contract values concentrated in a small number of very large accounts.

This page was written from 10 sources, 6 on domains other than afterquery.com.

1. 1.afterquery.com — afterquery.comvendor
2. 2.afterquery.com — productsvendor
3. 3.afterquery.com — leaderboardvendor
4. 4.afterquery.com — how afterquery helped nvidia hill climb gdpvalvendor
5. 5.ycombinator.com — afterquery
6. 6.siliconangle.com — ai training data startup afterquery nabs 30m investment
7. 7.techcrunch.com — afterquery reportedly becomes y combinators fastest ever uni
8. 8.sacra.com — afterquery
9. 9.trustpilot.com — afterquery.com
10. 10.hn.algolia.com — search