What Is an Applied Research Lab?

AfterQuery is an applied research lab curating data solutions to accelerate foundation model development. Credit: AfterQuery

Backed by Altos Ventures and Y Combinator, AfterQuery is an applied research lab that encodes human expertise into data to train reliable AI models

When Spencer Mateega and Carlos Georgescu set out to build financial AI agents, they hit a wall. The problem was not the code but the data.

Seeing generic web scrapes leaving models blind to real-world nuance, the two realised that smarter AI required better experience rather than more text.

Establishing AfterQuery in 2025 while completing their university studies, Spencer and Carlos shifted the focus from building end-user applications to solving the industry-wide data bottleneck.

The future of AI won’t be trained on more data, it will be trained on better thinking

AfterQuery

AfterQuery is an applied research lab dedicated to encoding expert human judgment into AI models.

As web-scale data reaches its limits, model performance is fundamentally bound by training data quality.

Understanding this, the organisation designs specialised datasets, reinforcement learning environments and evaluation frameworks that train reliable AI models.

How an applied research lab operates

Traditional academic institutions focus on theoretical breakthroughs, while enterprise data vendors label simple text. An applied research lab operates directly on the frontier of implementation.

AfterQuery treats data curation as a rigorous research discipline. The lab investigates how models learn, where they fail and how human intuition translates into algorithmic feedback loops.

By pairing software engineering with domain expertise, AfterQuery creates environments for models to practice reasoning. These systems allow AI to learn from error-recovery cycles and meet professional standards.

The lab designs its research around real-world application rather than abstract testing. This approach ensures AI systems perform reliably in high-stakes professional workflows.

Spencer Mateega co-founded AfterQuery along with Carlos Georgescu in 2025. Credit: Spencer Mateega/LinkedIn

Research areas driving data quality

To bridge general language fluency and professional intelligence, AfterQuery focuses on five main research pillars. Each pillar addresses a specific gap in current model development.

• Developing computer-use environments that teach AI agents to navigate software workflows and edge cases
• Building multimodal datasets that pair language with complex documents, interfaces and engineering charts
• Collaborating with security specialists to surface adversarial scenarios and construct guardrail data
• Designing curation methodologies that guarantee every example reflects expert human judgment
• Creating model evaluation benchmarks that measure whether reasoning steps match expert mental models.

Products built for model training

AfterQuery converts research insights into practical training tools for AI labs. Its product suite focuses on capturing human nuance, tool interaction and domain expertise across every stage of model development.

The lab builds tool-calling reinforcement learning (RL) environments using real APIs and Model Context Protocol (MCP) servers. These custom sandboxes teach software agents how to chain actions, call external services and recover from errors inside realistic workflows.

Computer-use and browser-use environments pair high-fidelity software interfaces with expert demonstration trajectories. Supervised fine-tuning (SFT) datasets supply high-quality prompt-response pairs and reasoning traces to establish foundational skills before RL begins.

Rubric and verifier-based RL combines expert-crafted criteria with automated verifiers to grade model outputs. Meanwhile, RLHF captures subtle human taste across thousands of expert comparison pairs.

To embed domain-specific judgment, deep research solutions draw on nearly 100,000 verified practitioners across medicine, law, finance and engineering. Systematic loss analyses identify precise model failure modes, while off-the-shelf datasets offer immediate access to pre-validated training data.

AfterQuery’s training data includes:

• Supervised Fine-Tuning (SFT)
• Reinforcement Learning + Rubrics
• Agent Environments (API/MCP)
• Computer Use Trajectories

Investors supporting the lab

In 2025, AfterQuery completed a US$30m Series A funding round led by Silicon Valley investment firm Altos Ventures. The funding round valued the company at US$300m.

Subsequent market developments in September 2026 saw the company reach a US$3.2bn valuation. This milestone made AfterQuery the fastest start-up in Y Combinator history to achieve unicorn status.

The rapid valuation growth reflects an annual recurring revenue run rate in excess of US$100m. This capital injection enables AfterQuery to scale its research infrastructure and expand its network of verified subject-matter experts.

The company receives backing from prominent venture capital firms across the technology sector. Early support from Y Combinator provided foundational momentum, while follow-on funding came from venture specialists including The Raine Group, BoxGroup and Latitude Capital.

Individual angel investors from leading AI organisations, including Google DeepMind, OpenAI, Anthropic and Meta, also support the lab. Their involvement brings specialised technical insight to complement AfterQuery’s financial backing.

Key partners

AfterQuery collaborates with hardware providers, AI platforms and academic research institutions to advance model capabilities. These include:

• NVIDIA: The enterprise software and GPU accelerated computing company utilises AfterQuery datasets to support the training of frontier models.
• Legora: A legal AI platform which partners with AfterQuery to build legal evaluations for research and diligence tasks.
• Thinking Machines Lab: A research collective which collaborates on advanced model evaluation and data methodologies.
• Academic Research Labs: Research institutions, including UC Berkeley, Stanford AI Lab and Allen Institute for AI, which partner on open dataset benchmarks.