🤖 AI CV Chatbot

RAG-based chatbot that answers questions about CV and projects using open-source LLMs and vector search.

🎯 Overview

This chatbot uses Retrieval-Augmented Generation (RAG) to provide accurate, context-aware answers based solely on your CV and project information. It prevents hallucination by grounding all responses in the provided data.

🏗️ Architecture

User Question
      ↓
FastAPI Backend
      ↓
Query Embedding (sentence-transformers)
      ↓
FAISS Vector Search
      ↓
Relevant Chunks Retrieved
      ↓
Prompt Template with Context
      ↓
Open-Source LLM (Mistral-7B via Ollama)
      ↓
Answer

🧩 Components

• LLM: Mistral-7B-Instruct (local inference via Ollama)
• Embedding Model: sentence-transformers/all-MiniLM-L6-v2
• Vector Database: FAISS (local, fast, free)
• Backend: FastAPI
• RAG Pipeline: Custom implementation with context retrieval

📁 Project Structure

ai-cv-chatbot/
│
├── data/
│   ├── cv.txt          # CV information
│   ├── projects.txt    # Project descriptions
│   └── skills.txt      # Skills and technologies
│
├── ingest.py           # Data ingestion → FAISS index
├── rag.py              # RAG retrieval + generation
├── llm.py              # Ollama LLM client
├── app.py              # FastAPI backend
│
├── requirements.txt
├── Dockerfile
└── README.md

🚀 Setup

Prerequisites

1. Python 3.11+
2. Ollama installed and running

   # Install Ollama from https://ollama.ai
   # Pull Mistral model
   ollama pull mistral:7b-instruct
   # Start Ollama server
   ollama serve

Installation

1. Clone and navigate to project

   cd cv_chatbot

2. Install dependencies

   pip install -r requirements.txt

3. Prepare your data

   • Edit data/cv.txt with your CV information
   • Edit data/projects.txt with your projects
   • Edit data/skills.txt with your skills

4. Ingest data into vector database

   python ingest.py

   This creates faiss_index.bin and metadata.txt.

5. Start the API server

   python app.py

   Server runs on http://localhost:8000

🌐 Publish (Free) on GitHub

This repo can be published for free as:

• Frontend: GitHub Pages (static hosting)
• Backend (FastAPI + Ollama): runs locally (GitHub Pages cannot run Python/LLMs)

1) Create a public GitHub repo and push

cd cv_chatbot
git init
git add .
git commit -m "Initial commit"
git branch -M main
git remote add origin https://github.com/<your-username>/<your-repo>.git
git push -u origin main

2) Enable GitHub Pages for the frontend

This repo includes a GitHub Actions workflow at .github/workflows/deploy-frontend.yml that builds the Vite app in frontend/ and deploys it to GitHub Pages automatically on every push to main.

In GitHub:

• Go to Settings → Pages
• Under Build and deployment, select GitHub Actions

After the workflow completes, your site will be available at:

• https://<your-username>.github.io/<your-repo>/

3) Point the frontend to a backend (optional)

By default the frontend uses http://127.0.0.1:8000. If you later host the backend publicly, set:

• Vite env var: VITE_API_URL (example: https://your-api.example.com)

📡 API Usage

Ask a Question

curl -X POST "http://localhost:8000/ask" \
  -H "Content-Type: application/json" \
  -d '{"question": "What programming languages do you know?"}'

Response:

{
  "answer": "Based on the provided context, I know Python, JavaScript, and SQL.",
  "sources": ["skills.txt", "cv.txt"],
  "chunks": [
    {
      "source": "skills.txt",
      "text": "Programming Languages: Python, JavaScript, SQL",
      "relevance_score": 0.15
    }
  ]
}

Health Check

curl http://localhost:8000/health

API Documentation

Visit http://localhost:8000/docs for interactive Swagger UI.

🐳 Docker

Build and Run

# Build image
docker build -t cv-chatbot .

# Run container
docker run -p 8000:8000 cv-chatbot

Note: For Docker, you'll need to run Ollama separately or use a Docker Compose setup.

🔧 Configuration

Change LLM Model

Edit llm.py:

DEFAULT_MODEL = "phi3:mini"  # or any Ollama model

Adjust Retrieval

Edit rag.py:

TOP_K = 5  # Retrieve more/fewer chunks

Change Embedding Model

Edit ingest.py and rag.py:

EMBEDDING_MODEL = "sentence-transformers/all-mpnet-base-v2"  # Larger, more accurate

🎓 Key Features

✅ No Hallucination: Answers only from provided context
✅ Open-Source: Uses free, local LLMs
✅ Fast: FAISS vector search is extremely fast
✅ Scalable: Easy to add more data sources
✅ Production-Ready: FastAPI, Docker, proper error handling

📝 Data Format

Data files should contain plain text, one sentence per line or paragraph format:

Your CV information here.
Multiple sentences are fine.
They will be chunked automatically.

🐛 Troubleshooting

"FAISS index not found"

Run python ingest.py first.

"Cannot connect to Ollama"

• Make sure Ollama is running: ollama serve
• Check if model is installed: ollama list
• Install model: ollama pull mistral:7b-instruct

"Model not found"

Install the model in Ollama:

ollama pull mistral:7b-instruct

📚 Technologies

• FastAPI: Modern Python web framework
• FAISS: Facebook AI Similarity Search
• sentence-transformers: State-of-the-art embeddings
• Ollama: Local LLM inference
• Mistral-7B: Open-source instruction-tuned LLM

🎯 Use Cases

• Personal portfolio chatbot
• CV/interview preparation
• Project showcase
• Knowledge base Q&A

📄 License

MIT

---

Built with ❤️ using RAG and open-source AI