I like building stuff — mostly backend systems, distributed pipelines, and lately things at the intersection of LLMs and memory.
Memo-grafter — LLM memory library that isn't just "dump everything into context and pray." Context that travels across chatbots, pulls only what's relevant, and stays token-efficient. Still cooking.
Checkout -> MemoGrfter Playground
Searchify — Built a search engine from scratch to understand how one actually works. Custom crawler, hybrid retrieval (BM25 + vector), RAG answer layer. Since I could only crawl so many links, I added a "critical insight" layer that tells you what the answers are probably missing. Constraint became a feature.
- Contributed to Docmost and Letta-code.
- Node.js · PostgreSQL · Redis · Elasticsearch · Docker · BullMQ · pgvector