Build a small retrieval-augmented (RAG) Q&A

The question

How can I create a create a chatbot and use it as a small RAG over my case studies?

pgvector allows PostgreSQL to store vectors... very handy
Anthropic does not provide its own embedding API, Voyageai is recommended for embedding. (i.e. You need Voyageai to create your vectors). OpenAI, on the other hand, does include embedding logic.
The amount of context in your embeddings greatly dictates the search results.
Embeddings are about semantic similarity, not reasoning.

2026-09-26 — Researched the RAG methodology; Started the search code... doc ingesting, chunking, vectorization, ask question, return highest similarity chunk.
2026-09-27 - Updated postgres to 17 in order to support pgvector, installed pgvector, created mini-rag database.
2026-09-28 - ~~Tweaked doc parsing to recognize my markdown sections so each chunk is a section, not just 50 chars.~~ Created the SQL to insert each section name, content, and vector. Finished the ingestion side of my mini_rag database.