Spencer Mateega and Carlos Georgescu’s AfterQuery started out as a college-era pitch to join Y Combinator. But the San Francisco startup quickly ran into a tough reality about AI training: models don’t really fail because they can’t reason—they fail because they haven’t seen enough real-world professional workflows.
The company shifted course, moving away from AI agents for finance. Instead, it focused on building high-quality, human-validated training data that actually captures the messy judgment calls and handoffs of real work.
In a short time, this focus brought rapid growth and a product philosophy centered on validation. They’ve carved out a spot in a pretty unpredictable market for human-derived data.
What AfterQuery does and why it matters
AfterQuery builds custom software systems that validate training data. They run human-generated examples through tough checks to land at a “Goldilocks” level of difficulty—tricky enough to push current AI models, but not impossible.
This is different from vendors who just try to scale up or use automated interviewing tools. By curating datasets that reflect real professional decision points, AfterQuery wants to help models learn faster and make those gains stick.
The company publishes research and runs post-training pipelines that show objective improvements when models use its data. Before clients even see the data, AfterQuery trains models on its datasets to show real, measurable gains.
This validation-first approach sets them apart from competitors who mostly chase volume or automation over quality instruction.
Founding story and pivot
Mateega and Georgescu launched AfterQuery after a last-minute Y Combinator application in college. Georgescu left school to go all-in on the startup, and the founders shifted their focus from a finance AI agent to the deeper problem of data quality.
They realized that AI’s big bottleneck wasn’t reasoning—it was the quality and relevance of training material, especially the kind that reflects how professionals actually work. That insight set the company on a path to produce and validate training data that fits real-world workflows.
Goldilocks: a validation-first data strategy
AfterQuery stands out with its end-to-end pipeline for data quality. The team builds custom validation software and applies strict checks, making sure each example falls in the Goldilocks range: challenging, but not overwhelming.
They also train models on their own data and benchmark results before showing anything to clients. The idea is to get repeatable, auditable improvements—not just one-off wins.
- Custom validation tools that enforce data quality standards
- Human-generated examples vetted through rigorous quality checks
- Transparent benchmarks and objective performance demonstrations
How training data is produced and validated
The workflow focuses on real-world professional tasks. They capture the nuance of judgment calls, handoffs, and decision points you’d actually see in an enterprise setting.
AfterQuery’s pipelines aim to produce datasets that show improvements on established benchmarks. They share research outputs to back up those gains. It’s not just about more data—it’s about data that teaches models to handle authentic, messy situations out in the wild.
Market context and investments
AfterQuery has grown fast, hitting a >$100 million annual revenue run rate within a year of its pivot. The team’s up to about 30 employees in San Francisco.
They raised a $30 million Series A at a $300 million valuation, led by Altos Ventures. The Raine Group, Y Combinator, and BoxGroup joined in, along with researchers from Anthropic, OpenAI, Google DeepMind, Meta’s Superintelligence Labs, and Microsoft.
The human-data market is booming, but it’s not exactly stable. Big breaches, like Mercor’s, show the risk of customer churn and reputational damage if quality or governance slip.
For AfterQuery, keeping up this pace will mean doubling down on rigorous data standards and making sure their validation pipelines hold up as demand from leading AI labs keeps rising.
Impact on AI labs and the broader ecosystem
For AI labs and enterprise buyers, AfterQuery’s model actually offers a pretty compelling path to real model improvements. The key is that these improvements are grounded in high-quality, auditable data.
Enterprises that want reliable AI might start to care a lot more about validated data pipelines and transparent benchmarking. They also want to show objective gains before rolling anything out.
With so many novelty tools flooding the market, a data-centric approach that puts quality, governance, and measurable outcomes first might just set the new standard. It feels like that’s the direction sustainable AI is heading, even if not everyone’s there yet.
Here is the source article for this story: This 23 Year-Old’s New AI Data Company Has Already Hit A $100 Million Run Rate