Company profile · 4 min read
AfterQuery turns expert decisions and domain work into training data for frontier models
AfterQuery coordinates expert-led datasets, benchmarks and agent environments for AI teams working on specialized professional tasks.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
AfterQuery is a fit for an AI lab that needs expert-generated data for difficult professional work, not another generic labeling vendor. Its pitch is that models need the decisions, tradeoffs and context behind an answer—not only more completed outputs.
What it does
AfterQuery says it works with domain experts to turn real-world work into training data for frontier models. Its public materials describe supervised fine-tuning pairs, human-led reinforcement-learning data, custom benchmarks, multimodal data and computer-use trajectories. The examples span coding, finance, legal work and healthcare. AfterQuery homepage YC launch
That is a more specific wedge than “we label your data.” A team building a finance agent, legal workflow or developer tool may need examples of how a qualified person evaluates an ambiguous case, handles an exception and explains a decision. AfterQuery’s public offer is to assemble the relevant experts and coordinate the dataset, so the model team can focus on training and evaluation.
Why it may fit
The public site is strongest when the failure is reasoning in a domain rather than broad language coverage. It also advertises research and a marketplace-style operating layer, while its careers page shows roles spanning expert operations, platform, RL environments and data privacy. AfterQuery careers That combination suggests a company trying to own the supply and production process, not merely resell annotator hours.
The founder context is relevant to a technical buyer. The YC profile describes Spencer Mateega’s software, finance and statistics background and Carlos Georgescu’s software engineering and prior acquired ed-tech experience. Those are public professional facts, not proof that a proposed dataset will meet a model team’s quality bar.
What to ask before committing
The tradeoff is custom depth versus procurement and repeatability. Ask how experts are qualified for the exact task, how disagreements are adjudicated, how sensitive material is handled, how reasoning is represented without leaking private information, and whether the resulting labels can be regenerated consistently. The public pages do not list standard pricing or a fixed turnaround, so a buyer should request a bounded sample with acceptance criteria before placing a large order.
My editorial take
I would use AfterQuery when a model’s benchmark failure is visibly tied to domain judgment—especially when internal experts are too scarce to build the dataset themselves. I would not hire it simply to expand a broad web corpus. The useful first engagement is one narrow task with known failure examples, named expert requirements and a clear test set; if the sample does not improve the target behavior, more annotation will not rescue the project.
Quick facts
| Field | Sourced detail |
|---|---|
| Buyer | Frontier model teams and specialized AI product teams |
| Product | Expert-led datasets, benchmarks and agent environments |
| Data types | SFT, RL data, custom benchmarks and computer-use examples |
| Commercial signal | Contact-led; no public pricing in the checked sources |
| Main fit question | Is the bottleneck domain reasoning rather than raw data volume? |
Sources checked
AfterQuery’s YC profile, company homepage, launch page and careers page were checked on 2026-09-19.
Cohort context
AfterQuery is listed in Winter 2025. In our 2026-09-18 directory snapshot, 104 of 165 listed companies in that cohort have YC’s primary industry label B2B (63.0%). This is a current-directory comparison, not an original intake count or a performance ranking. Nine-cohort dataset.
Public website snapshot
Observed 2026-09-19T16:19:21.028Z in raw homepage HTML. This records visible metadata and advertised links, not agent execution or product quality.
| Signal | Homepage observation |
|---|---|
| Product description metadata | Observed |
| Canonical link | Observed |
| H1 or H2 heading | Observed |
| Typed structured data | Observed |
| Docs/developer link | Not observed in this response |
| Pricing link | Not observed in this response |
| llms.txt link | Not observed in this response |
| Markdown alternate | Not observed in this response |
Public observations · Collection method. Missing links here do not establish that a capability or file is absent elsewhere.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗