How it works
The chat on the home page is retrieval-augmented generation (RAG). My notes live as Markdown files. They are split into chunks and turned into vectors. When you ask something, the closest chunks are found and handed to a language model, which answers from them. Click any step to learn more, or press play.
Top row: indexing, runs once per content changeRows 2–3: answering, runs per question
Indexing · step 1 of 10
Markdown notes
Everything the assistant knows comes from plain .md files in the content folder: bio, projects, experience. Adding knowledge means dropping in a new file, with no database or retraining. Right now: about.md.
content/*.md
Try the retrieval step
Type a question to see which chunks the model would receive and how similar each one is.