RAG pipeline
Chat with your documents: chunking, embeddings, retrieval, reranking and how to know it works.
What is on this board
Parts
- Reader User
- Chat Web app
- Answers Service · Python
- Checks Guardrails / evals
- Embed question Embeddings · OpenAI
- Passages Vector DB · Pinecone
- Rerank ML model
- Model LLM · Claude
- Same question Cache
- Your documents Object storage · S3
- Read PDFs OCR / document AI
- Chunk and embed Batch job
- Evals nightly Scheduled job
- What moved the needle Data table
How they connect
- Reader → Chat: asks
- Chat → Answers: question
- Answers → Same question: seen it?
- Answers → Embed question: vector
- Answers → Passages: closest 20
- Answers → Rerank: best 5
- Answers → Model: answer with sources
- Answers → Checks: safe, and from the sources?
- Your documents → Read PDFs: new document
- Read PDFs → Chunk and embed: text
- Chunk and embed → Passages: store
- Evals nightly → Answers: known questions
