Research — pushing the frontier of AI-assisted development.
Our work on secure codebase indexing, semantic search, reinforcement learning for code agents, and multi-agent collaboration systems.
View publications
We're an applied research team focused on building AI that helps developers ship ambitious software faster and more reliably than ever before. Our research spans model training, systems engineering, and human-computer interaction.
Research areas
Codebase indexing
Custom embedding models that index entire codebases with high recall. Agents get instant access to relevant code, documentation, and context across millions of lines.
Semantic search for code
State-of-the-art semantic search that understands code semantics, not just syntax. Ask questions in natural language and find exactly what you need.
Reinforcement learning for agents
Training coding agents with RL on real-world scenarios. Our agents learn from millions of successful coding sessions to make better decisions.
Multi-agent collaboration
Systems that enable multiple AI agents to work together on complex tasks, dividing work, sharing context, and collaborating in parallel.
Tab prediction models
Specialized transformer models trained for sub-100ms code completion. Our Tab model predicts multi-line edits with industry-leading accuracy.
Privacy-preserving AI
Techniques for running AI models without compromising user privacy. Including zero data retention, on-device processing, and differentially private training.
Publications
Secure codebase indexing at scale
How Zod securely indexes millions of lines of code across thousands of repositories without compromising privacy or performance.
Semantic search for code
Building high-recall semantic search across large codebases using custom-trained embedding models and retrieval-augmented generation.
Reinforcement learning for code agents
Training coding agents with RL on millions of real-world scenarios. Our approach achieves state-of-the-art results on SWE-Bench.
Multi-agent collaboration systems
Enabling multiple agents to work together on complex tasks through shared context, task decomposition, and parallel execution.
Conferences & appearances
Our team regularly presents at top ML and software engineering conferences worldwide. Meet us at our next event.
See upcoming eventsJoin our research team.
We're hiring research scientists and ML engineers to push the frontier of AI-assisted development.
View open positions