📚 RAG Module — Retrieval-Augmented Generation¶

From fundamentals to production design, plus a small working RAG app.

Overview¶

Retrieval-Augmented Generation (RAG) improves LLM answers by first retrieving relevant passages from a knowledge base and putting them in the prompt. The model answers from current, domain-specific, citable sources instead of only its training data.

User Query → Retrieve (search the index) → Augment (add chunks to the prompt) → Generate (LLM answer + citations)

Contents¶

#	Document	Description
1	RAG Fundamentals	Pipeline stages, chunking, embeddings, vector stores, evaluation, common failures
2	LM Studio Integration	Running a small local model (Gemma, Qwen, Llama...) behind an OpenAI-compatible API
3	Fine-Tuning Guide	When and how to fine-tune embeddings, rerankers and the generator
4	RAG Interview Questions	14 Senior/Staff-level questions: hybrid search, contextual retrieval, GraphRAG, evaluation, security
5	Code Base Design	How the implementation is structured: SOLID, patterns, error handling, tests
6	Low-Level Design	Class and sequence diagrams, data models, API contract, pgvector schema
7	High-Level Design	Production architecture, capacity estimates, latency, cost model

Implementation¶

implementation/: working Python RAG chatbot (FastAPI, ChromaDB, sentence-transformers, LM Studio)
test_pipeline.py: end-to-end check over the sample corpus in data/ (no LLM server needed)

Quick Start¶

cd ai-engineering/rag
python -m venv .venv && source .venv/bin/activate
pip install -r implementation/requirements.txt

python test_pipeline.py                                   # index data/*.md into a temp store, run checks
python implementation/main.py --index --docs ./data       # index the sample corpus
python implementation/main.py --query "What is RAG and how does it work?"   # needs LM Studio running