Current projects

A chat assistant that answers real questions about what I do — grounded in my own writing and project notes rather than free-associating. The content is chunked, embedded as vectors, and retrieved by semantic similarity, so answers stay anchored to source material instead of being made up (retrieval-augmented generation).

It doubles as a live demo. The same pattern — embeddings, a vector store, retrieval, an LLM on top — is exactly what I'd build for a client who wants an assistant grounded in their own documents rather than the open internet.