Resume Chatbot (LangChain + FastAPI + Streamlit)

Conversational Q&A over a resume using Retrieval-Augmented Generation (RAG). This repo provides:

• FastAPI backend with chat + session APIs
• LangChain pipeline with FAISS vector store
• Conversational memory per session
• Streamlit UI with an Index Rebuild button

🎨 Features

• 💬 Chat interface with conversation history
• 📄 Source document viewer
• 🗑️ Clear chat history
• 📱 Responsive design
• 🎨 Custom theming

🎯 Quick Start

Prerequisites:

• Python 3.10+ (3.12 tested)

Setup and run locally:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# Copy your resume to the repo root as resume.pdf (or set RESUME_PATH)

# Start Streamlit UI (preferred)
python -m streamlit run streamlit_app.py

# Or start the FastAPI server directly
python main.py

Tip: If streamlit command isn’t found, always use python -m streamlit ....

🏗️ Architecture

• app/core/app.py – FastAPI app factory, CORS, routes
• app/services/rag_service.py – Loads resume, builds vector store, exposes query() and rebuild()
• resume_loader.py – Loads and splits the resume PDF into chunks
• rag_chain.py – Builds the LangChain RAG pipeline (LLM + retriever + prompts)
• app/services/memory_service.py – In-memory per-session chat history
• streamlit_app.py – Streamlit interface (chat UI + Rebuild Index)

Data flow:

1. Resume PDF -> chunked -> embedded -> FAISS index
2. User question + chat history -> RAG chain -> answer (+ optional sources)

📁 File Tree (major parts)

.
├── main.py                      # Run FastAPI server
├── streamlit_app.py             # Run Streamlit UI
├── resume_loader.py
├── rag_chain.py
├── email_sender.py
├── app/
│   ├── core/
│   │   ├── app.py               # create_app()
│   │   └── config.py            # pydantic settings
│   ├── api/
│   │   └── routes/
│   │       ├── chat.py          # POST /api/v1/chat
│   │       ├── sessions.py      # GET/DELETE session endpoints
│   │       └── health.py        # GET /health
│   ├── models/schemas.py        # Pydantic models
│   └── services/
│       ├── rag_service.py       # RAGService
│       └── memory_service.py    # ChatMemoryService
└── .env                         # local secrets (gitignored)

💡 Configuration (.env)

This project reads settings from .env in the repository root. Never commit real secrets.

Example .env template:

# Required
OPENAI_API_KEY="sk-..."
RESUME_PATH="resume.pdf"           # Path to your resume PDF

# Vector store (FAISS)
VECTORSTORE_PATH=.faiss_index       # Directory for FAISS index

# LLM / Embeddings
LLM_MODEL=gpt-4o-mini
LLM_TEMPERATURE=0.2
USE_LOCAL_EMBEDDINGS=false          # Set true only if configured

# Retrieval tuning
RETRIEVER_K=4
RETRIEVER_FETCH_K=12
MMR_LAMBDA=0.7

SESSION_TIMEOUT_MINUTES=30  # Auto-cleanup inactive sessions after 30 minutes

CLEANUP_INTERVAL_SECONDS=180  # Run cleanup every 180 seconds

Additional configurable settings live in app/core/config.py (with safe defaults): host/port, CORS, memory limits, etc.

Security note: rotate any leaked keys and keep .env in .gitignore.

🧑‍💻 Running

Streamlit UI:

python -m streamlit run streamlit_app.py

FastAPI server:

python main.py
# or
uvicorn main:app --reload --host 0.0.0.0 --port 8000

By default, the API listens on http://localhost:8000.

API Endpoints

• GET /health

  • Returns service status and version.

• POST /api/v1/chat

  • Body (JSON):

     {
     	"question": "What programming languages does this candidate know?",
     	"session_id": "user123"
     }

  • Response (JSON):

     { "response": "... answer ..." }

• GET /api/v1/sessions/{session_id}/history

  • Returns an array of {question, answer} entries and a count.

• DELETE /api/v1/sessions/{session_id}

  • Clears the session’s chat history.

• GET /api/v1/sessions

  • Lists active sessions and message counts.

Example curl:

curl -s http://localhost:8000/health | jq

curl -s -X POST http://localhost:8000/api/v1/chat \
	-H 'Content-Type: application/json' \
	-d '{"question":"Summarize experience","session_id":"demo"}' | jq

💻 Streamlit UI

Run the app and open the chat interface:

python -m streamlit run streamlit_app.py

Sidebar features:

• Model and Session ID display
• Clear Chat History
• Index section showing VECTORSTORE_PATH and whether FAISS exists
• "Rebuild Index" button that deletes and rebuilds the FAISS index and clears chat

Automatic refresh:

• On startup, the backend computes a SHA256 of your resume. If it changed since the last run, the index is automatically rebuilt.

📇 Index Management (FAISS)

• Path is controlled by VECTORSTORE_PATH (default: .faiss_index).
• To force a rebuild:
  • Use the Streamlit sidebar: "Rebuild Index"
  • Or delete the directory manually and restart:

     rm -rf .faiss_index

Metadata:

• The service writes .faiss_index/meta.json with the resume hash and docs count for change detection.

🐛 Troubleshooting

• Streamlit command not found:

   python -m streamlit run streamlit_app.py

• "Resume not found" on startup:

  • Ensure RESUME_PATH points to an existing PDF file
  • In Codespaces/Cloud, make sure the file is present in the workspace

• Old index after updating resume:

  • Click "Rebuild Index" in Streamlit, or remove .faiss_index/ directory

• OpenAI auth errors:

  • Set OPENAI_API_KEY in .env or environment

• Port already in use:

  • Change port in .env (PORT) or run uvicorn ... --port 8001

Deployments

General guidance:

• Include your resume PDF in the repo (or mount it)
• Set secrets (API key, paths) in the hosting platform’s secrets UI
• The first launch may take time to build the index

Notes & Limits

• Chat memory is in-process and ephemeral. A restart clears it.
• Email notifications (email_sender.py) are optional; configure SMTP only if you need them.