🤖 AI-Powered Parenting Assistant

📌 Overview

The AI-Powered Parenting Assistant is an intelligent chatbot designed to provide safe, reliable, and empathetic responses to parenting-related questions. Built using LangChain, LLM Guard, and FastAPI, the system leverages Retrieval-Augmented Generation (RAG) to enhance its responses with relevant knowledge from curated sources.

The chatbot also integrates advanced content filtering to ensure the safety and appropriateness of responses, making it a powerful demonstration of AI engineering best practices.

---

🚀 Features

✅ Retrieval-Augmented Generation (RAG) System

• Uses FAISS vector search combined with BM25 retrieval for accurate contextual knowledge retrieval.
• Implements OpenVINO Reranker for document ranking and relevance filtering.

🛡 AI Safety with LLM Guard

• Input Filtering: Blocks toxic, political, religious, and self-harm topics using llm_guard.
• Output Validation: Scans model responses for language safety, sensitivity, and relevance.
• Routing Mechanism: Ensures inappropriate queries are handled gracefully.

⚡ High-Performance AI Pipeline

• LangChain Pipelines: Modular runnable chains for seamless data flow.
• Prompt Engineering: Optimized few-shot prompting for structured AI outputs.
• FastAPI Backend: Fully asynchronous REST API for real-time chatbot interactions.
• Langfuse Monitoring: Enables real-time LLM performance tracking.

🔥 Job-Worthy Tech Stack

✅ AI & LLMs: OpenAI GPT Models, LangChain, HuggingFace Embeddings
✅ Data Retrieval: FAISS, BM25, OpenVINO Reranker
✅ Safety: LLM Guard (Input & Output Filtering)
✅ Backend API: FastAPI, Pydantic, Loguru
✅ Monitoring: Langfuse Callbacks
✅ Dockerized Pipeline: Multi-stage AI data processing stack

---

📁 Project Structure

📂 src/
 ├── core/
 │   ├── chatbot.py       # Main AI chatbot pipeline (LLM + Retrieval + Guard)
 │   ├── embedding.py     # FAISS/BM25 index generation
 │   ├── evaluation.py    # AI performance and accuracy checks
 │   ├── filters.py       # LLM Guard configurations for input/output scanning
 │── config.py            # Application settings & environment variables
 │
 ├── api/
 │   ├── routers/
 │   │   └── chat.py      # FastAPI routes for chatbot integration
 │   │── __init__.py      # API initializer
 │   └── main.py          # FastAPI app initialization
 │
 ├── data/
 │   ├── clean_data.py    # Data preprocessing scripts
 │
 ├── monitoring/
 │   └── monitoring.py    # Langfuse monitoring setup
 │
 ├── Dockerfile           # Containerization setup
 ├── docker-compose.yml   # Multi-container deployment setup
 ├── requirements.txt     # Python dependencies
 ├── README.md            # Project documentation (this file)

---

🏗 Setup & Installation

1️⃣ Clone the Repository

git clone https://github.com/amine-akrout/parenting_assistant_rag.git
cd parenting_assistant_rag

2️⃣ Create Virtual Environment & Install Dependencies

python -m venv venv
source venv/bin/activate  # On Windows use `venv\Scripts\activate`
pip install -r requirements.txt

3️⃣ Set Up Environment Variables

Create a .env file in the root directory:

POSTGRES_USER=your_db_user
POSTGRES_PASSWORD=your_db_password
POSTGRES_DB=your_db_name
OPENAI_API_KEY=your-openai-key
LANGFUSE_PUBLIC_KEY=your-langfuse-public-key
LANGFUSE_SECRET_KEY=your-langfuse-secret-key

4️⃣ Run the API Server

uvicorn src.api.main:app --host 0.0.0.0 --port 8000

5️⃣ Test the API

Open Swagger UI to test endpoints:
📌 http://localhost:8000/docs

---

🐳 Dockerized Workflow

This project includes a Dockerized data pipeline that prepares, indexes, and serves the chatbot API.

1️⃣ Start All Services

docker-compose up --build

2️⃣ Stop All Services

docker-compose down

Docker Services Overview

• langfuse-server → Logs chatbot requests/responses for monitoring
• db → PostgreSQL database for logging and indexing
• clean-data → Prepares and cleans datasets before indexing
• preprocess → Creates FAISS/BM25 indexes for fast retrieval
• chatbot-api → Runs the FastAPI chatbot backend

---

📈 Performance Monitoring (Langfuse)

Enable Langfuse for real-time monitoring:

📌 http://localhost:3000

View logs on Langfuse Dashboard.

---

🧪 Testing

Run unit tests using pytest:

pytest tests/

---

🎯 Future Enhancements

✅ Add multi-turn conversations (memory support)
✅ Implement whisper model for speech-to-text
✅ Add more retrieval sources (vector & hybrid search)
✅ Integrate more LLM Guard features (e.g., sentiment analysis)

---

🤝 Contributing

Feel free to open an issue or submit a pull request.

---