jasma

Landing Page

Inspiration

We wanted to eliminate repetitive bug fixing and create a shared memory system that allows developers and coding agents to instantly benefit from solutions that have already been proven.

What it does

Jasma captures bug fixes, validates them, and stores them in a shared database. When a developer encounters a similar issue, the system retrieves and applies prior fixes automatically or uses a generative agent to propose new ones, then validates the outcome.

How we built it

We built an MCP server with Python, FastAPI, and TypeScript that integrates with Supabase for storage and pgvector for semantic search. Next.js and Vercel were used for the frontend, Cloudflare for routing, and EC2 for compute tasks. Selenium ChromeDriver was used for data scraping to supplement the bugfix knowledge base.

Challenges we ran into

Ensuring fixes were applied safely without overwriting valid code was difficult. Designing the branching and snapshot logic to allow validation and rollback required careful workflow planning. Integrating semantic retrieval with structured validation also presented challenges.

Accomplishments that we're proud of

We created a functioning system that can snapshot a repo, retrieve or generate fixes, validate them, and update a shared knowledge base. The entire pipeline was completed end-to-end during the hackathon.

What we learned

We learned how to combine generative AI with MCP and semantic search to build a reliable coding assistant. We also gained experience with Supabase’s pgvector, fast prototyping across multiple frameworks, and coordinating real-time validation.

What's next for Jasma

We plan to expand coverage to documentation lookup, integrate more robust validation frameworks, and improve cross-agent interoperability so that fixes are immediately usable across different developer tools and environments.