I build LLM systems that have to be right, not just fluent: agents that plan with structured outputs, retrieval that stays grounded in real data, and the evaluation harnesses that prove a change actually made things better.
- Software Development Engineer (AI) Intern at StyleMatch: building the LLM decision engine behind conversational product search, LLM catalog enrichment, hybrid semantic search, and the evaluation framework that keeps it all honest.
- Building in public: open-source AI tools, one at a time. First up:
dsa-coach-mcp.
| Project | What it does | Stack |
|---|---|---|
| dsa-coach-mcp | MCP server that coaches DSA practice on any platform: FSRS revision fitted to your own reviews, a mistake journal, an in-chat MCP Apps dashboard, and agent evals that grade how any model uses the tools. On PyPI and the official MCP Registry | Python · MCP · FSRS · SQLite |
| AskTube | Chrome extension that answers questions from YouTube videos with RAG | FastAPI · LangChain · RAG |
| Expense-Tracker-MCP-Server | MCP server for tracking and summarising expenses from any MCP client | Python · MCP · SQLite |
| ExplainIT | Learning assistant that explains any topic at beginner, technical or code level | Python · Groq · Prompting |
| Ai-Web-Scraper | Scraper for dynamic sites with LLM parsing and summarisation | Selenium · Ollama |
1 merged · 2 in review across 2 projects
pandas-dev/pandas ⭐ 50.6k
sqlfluff/sqlfluff ⭐ 10k
- RF03: don't unqualify correlated references in DML subqueries · 🟢 open
- Report relative paths when linting with no path argument · 🟢 open
Fun fact: I debug in Python and chat in German 🇩🇪