At a glance
Status: In progress · Stack: Python, Ollama, SQLite, FastAPI · Code: GitHub
Problem
I have hundreds of notes and PDFs, and keyword search doesn’t find what I mean.
Approach
- Embed notes with a small local embedding model
- Store vectors in SQLite
- Answer questions with an 8B model (see Running LLMs locally)
Result
Add screenshots, metrics or a demo link here.
Lessons
- Chunking mattered more than the model choice
- Evaluation came too late. Read Evaluating RAG pipelines