Research Engineer

We want to build agents that autonomously validate code changes. Today that looks like AI that reviews pull requests in GitHub, catching bugs and enforcing standards. We’re reviewing close to 1B lines of code a month now for over 3,000 companies.

Problems we’re excited about Coding standards can be idiosyncratic and are often poorly documented; can we build agents that learn them through osmosis like a new hire might? Can we identify for each customer what types of PR feedback they do and don’t care about, perhaps using some sample efficient RL, in order to increase signal-to-noise ratio? Some bugs are best caught by running the code, potentially against discerning AI-generated E2E tests.

Can we autonomously deploy feature branches and use agents to parallel try to break the application to detect bugs? Trajectory 7,000+ customers Raised $30M from Benchmark, YC, Paul Graham, Initialized Team We have assembled a small, talent dense team who have scaled critical functions at companies like Stripe, Google, Figma, LinkedIn, etc.

Qualifications

BS in Computer Science or equivalent Some research experience, ideally with ML/LMs/Agents Strong programming skills and good product intuition Responsibilities Experiment with and apply recent advancements in agents/LMs etc. to improve the performance and capabilities of our products Example: you might study multi-agent architectures, prototype and evaluate a multi-agent code review workflow, and then work with a team to integrate successful prototypes into production systems Stay current with the latest research in LLMs, information retrieval, and developer tooling We want to build agents that autonomously validate code changes. Today that looks like AI that reviews pull requests in GitHub, catching bugs and enforcing standards. We’re reviewing close to 1B lines of code a month now for over 3,000 companies.

Problems we’re excited about

• Coding standards can be idiosyncratic and are often poorly documented; can we build agents that learn them through osmosis like a new hire might?
• Can we identify for each customer what types of PR feedback they do and don’t care about, perhaps using some sample efficient RL, in order to increase signal-to-noise ratio?
• Some bugs are best caught by running the code, potentially against discerning AI-generated E2E tests. Can we autonomously deploy feature branches and use agents to parallel try to break the application to detect bugs? Trajectory - 7,000+ customers
• Raised $30M from Benchmark, YC, Paul Graham, Initialized Team
• We have assembled a small, talent dense team who have scaled critical functions at companies like Stripe, Google, Figma, LinkedIn, etc.

Qualifications

• BS in Computer Science or equivalent
• Some research experience, ideally with ML/LMs/Agents
• Strong programming skills and good product intuition Responsibilities
• Experiment with and apply recent advancements in agents/LMs etc. to improve the performance and capabilities of our products
• Example: you might study multi-agent architectures, prototype and evaluate a multi-agent code review workflow, and then work with a team to integrate successful prototypes into production systems
• Stay current with the latest research in LLMs, information retrieval, and developer tooling