NAVATAR-Helper

NAVATAR-Helper is an AI-driven chatbot developed to assist with health-related NEET (Not in Education, Employment, or Training) queries. It leverages the Retrieval-Augmented Generation (RAG) framework to provide accurate, bilingual answers in Norwegian and English based on verified sources, while minimizing hallucinations.

Overview

• Project Goal: Develop a fully working MVP of a RAG-based chatbot that delivers accurate, sourced & verified answers to NEET-related queries while minimizing hallucinations.
• Technology Stack:
  • Backend: Python, LangChain, custom LLMs.
  • Frontend: Streamlit for user interface.
  • Embedding Model: Alibaba-NLP/gte-multilingual-base for question vectorization.
  • Vector Database: Milvus for storing and retrieving vectorized data.
  • LLM: norallm/normistral-7b-warm-instruct for generating responses.
  • GPU Server: Nvidia GeForce GTX 1080 Ti (11GB VRAM).

Key Features

• RAG Framework: Integrates retrieval and generation for precise answers.
• Bilingual Support: Handles both Norwegian and English queries.
• Source Attribution: Provides references for each response.
• Reduced Hallucinations: Strict use of retrieved, relevant data ensures factual outputs.
• Extensive Testing: Both RAG and non-RAG approaches for performance comparison

Architecture

1. User Input: Processes user questions via a Streamlit-based frontend.
2. Language Detection: Determines the input language (Norwegian or English).
3. Embedding: Converts questions into vectors using Alibaba-NLP/gte-multilingual-base.
4. Context Retrieval: Utilizes Milvus to find relevant document chunks.
5. Response Generation: Uses norallm/normistral-7b-warm-instruct to produce context-informed answers.
6. Attribution: Displays sources for user verification.

Installation & Setup:

• Clone the repository:

  git clone https://github.com/sirin-koca/NAVATAR-Helper.git
  cd NAVATAR-Helper
  

1. Create virtual env: python3 -m venv venv

2. Activate:

   • Linux/macOS: source venv/bin/activate
   • Windows: venv\Scripts\activate

3. Install deps: pip install -r requirements.txt

4. Run Services:

• HTTP Server: python3 -m http.server 8503 (in Server folder)
• Client: streamlit run Client.py --server.port 8080 (in Client folder)
• Backend: python3 Server.py (in Server folder)

5. Access:

• Open: rag2.cs.oslomet.no (requires OsloMet network)
• Ensure all 3 services run simultaneously.

Backend scripts:

• popDB.py: Prepares the vector database with embeddings.
• setupLLM.py: Initializes the LLM for use.
• ragMain.py: Main interaction handler for user queries.

Prerequisites

• Python 3.8 or higher
• PyCharm or any Python IDE
• A virtual environment for dependency management
• Access to a GPU server (e.g., Nvidia GeForce GTX 1080 Ti or higher)
• Git for version control

Team Members: Sirin, Rafey, Younes, Morten, Valerie

---

DATA3750 Applied AI & Data Science | Group Project | OsloMet H2024 ©

---