What It Does

This recipe turns a pile of personal notes into a queryable “second brain”. You point it at an Obsidian vault (or exported Notion pages), chunk and embed the notes into a vector database, and then ask questions in plain language — “what did I decide about the pricing model?” — getting answers grounded in your own notes with links back to the source. Every new note you add makes the assistant more knowledgeable over time.

The Stack

• Your notes — a local Obsidian vault of markdown, or Notion pages exported to markdown.
• Embeddings — Jina AI Search Foundation offers free embedding + reranker API keys; many builders report indexing dozens of notes for a few cents.
• Qdrant Cloud Free Tier — a managed vector store to hold the chunk embeddings and run similarity + payload-filtered search.
• An LLM for synthesis — a free/BYOK chat model such as Mistral AI (La Plateforme) generates the final grounded answer.
• A RAG framework — LlamaIndex or LangChain to wire the ingest → embed → retrieve → answer pipeline.

Step-by-Step Outline

1. Ingest — walk the vault, split each note into overlapping chunks, and keep the source path as metadata.
2. Embed + store — embed each chunk and upsert into a Qdrant collection with the note path in the payload.
3. Retrieve — on a question, embed the query, pull top-k chunks (optionally rerank), and assemble context.
4. Answer with citations — prompt the LLM with the retrieved chunks and require it to cite the source note paths.
5. Keep it fresh — re-run ingest on changed files (or on a schedule) so new notes are searchable.

Source

“Build Your Own AI Second Brain in a Weekend” — architecture walkthrough for a personal notes RAG assistant (see also the open-source Second Brain AI Assistant course)