codeset - Building large-scale code datasets to train code agents

Our Solutions

Train code agents on large-scale datasets in a RL gym with real-world coding tasks.

Evaluate agents capabilities with Codeset's reliable benchmarks.

Create bespoke datasets from your internal codebase, tailored to your needs.

Leverage our platform to monitor the training and evaluation of your code agents on your datasets.

Co-Founder

Expert in machine learning on code, creator of RepairBench. Interned at Microsoft.

Co-Founder

Expert in software reliability, data mining, and IaC. Published at all top 3 SWE conferences (ICSE, FSE, ASE).

Co-creators of GitBug-Actions, a tool for generating executable and reproducible datasets to advance AI-driven software development.

Ready to elevate your AI models with high-quality code datasets? Let's collaborate on your next breakthrough.