Repository structure

This repo contains three layers of work:

• basic_examples/
  Small, single-file examples used to learn individual concepts (async, prompts, etc.).
  Think: quick experiments / scratchpad (not production).

• mini_projects/
  Day-by-day mini projects (p01…p09) that demonstrate one concept end-to-end
  (e.g., async API calling, LCEL, RAG baseline, MMR, hybrid retrieval + rerank, LangGraph workflow, MCP tools).

• src/app/ Main project (Capstone) The interview-ready application: a production-shaped RAG API built with FastAPI + LangGraph. This is the primary deliverable and the best entry point for reviewers.

Capstone project (src/app)

The capstone is a FastAPI service that exposes:

• GET /health — health check
• POST /query — grounded RAG Q&A

Internally it runs a LangGraph workflow with conditional routing: retrieve → answer → verify
If the verifier confidence is low, the graph retries retrieval up to a configured limit, otherwise it returns the final answer.

Key features:

• RAG end-to-end (load → chunk → embed → store → retrieve → generate)
• strict grounding + "I don't know" fallback when docs do not support an answer
• structured verification (confidence + ok/retry decision)
• logging + env-based configuration (no secrets in code)
• unit + integration tests

How it works (data flow)

Client -> POST /query -> LangGraph workflow: 1) retrieve: fetch top-k chunks from vector store 2) answer: generate answer using ONLY retrieved context 3) verify: score support/confidence and decide ok/retry -> Response: answer + sources + confidence + iterations

Running the capstone

1) Setup

python -m venv .venv
# activate venv
pip install -r requirements.txt

Create .env from .env.example and set OPENAI_API_KEY.

2) Run API

uvicorn src.app.main:app --reload

3) Example request

curl -X POST "http://127.0.0.1:8000/query" \
  -H "Content-Type: application/json" \
  -d "{\"question\":\"What is LCEL?\"}"