🧠 Arabic RAG Chatbot (Bosala AI)

Bosala AI is an Arabic Retrieval-Augmented Generation (RAG) chatbot that answers Arabic questions by retrieving relevant content from a vector database and generating responses using Google Gemini. The project supports both a web interface and command-line tools, is fully Dockerized, and is designed to be easy to use even for users with no prior ML or backend experience.

---

What This Project Does

• Stores Arabic documents as semantic vectors (embeddings) in Qdrant collection.
• Searches for the most relevant content when a user asks a question
• Sends the retrieved context to Gemini to generate a high-quality Arabic answer
• Tracks usage and performance through an analytics dashboard
• Provides user accounts, chat history, and admin controls via a web app
• Allow administrators to direct the compass to any knowledge base, through the ingest interface that processes texts and stores them in the database

---

Key Features

🇵🇸 Arabic-first RAG pipeline

🤖 Gemini 2.5 Flash for answer generation

📦 Qdrant vector database for semantic search

🌐 Django web application

• User registration & login
• Chat and Ingest UI
• Admin & analytics dashboard

🐳 Fully Dockerized

🖥️ CLI tools for ingestion, chat, and evaluation

---

High-Level Architecture

User Question → Embedding Model → Qdrant (Vector Search) → Relevant Context → Gemini (Answer Generation) → Final Arabic Answer

---

Main Components

• ragchat/ Core RAG logic: embeddings, retrieval, generation, evaluation, CLI tools

• backend/ Django web app:

  • UI
  • User authentication
  • Analytics dashboard
  • API endpoints

• data/ Stores raw data, processed data, and Qdrant vector storage

• load_testing/ Uses locust script to test the RAG pipeline

• Docker & Docker Compose Runs everything with a single command

---

Requirements

• Docker
• Docker Compose
• Google Gemini API key

No Python, Django, or ML knowledge required to run ')

---

Setup (One Time)

1. Clone the repository

   git clone https://github.com/HalaKhalifa/rag_arabic_chatbot.git
   cd rag_arabic_chatbot
   

2. Create a .env file

   GEMINI_API_KEY=your_gemini_api_key_here
   API_SECRET=random_secret
   EMB_MODEL=abdulrahman-nuzha/intfloat-multilingual-e5-large-arabic-fp16
   GEN_MODEL=models/gemini-2.5-flash
   GEN_MAX_NEW_TOKENS=512
   GEN_TEMPERATURE=0.4
   TOP_K=5
   

3. Running the Project

   a. Build everything

   docker compose build
   

   b. Start all services

   docker compose up
   

This will start:

• Qdrant (vector database)
• Django backend
• RAG services

---

Technologies Used

• Google Gemini
• Qdrant
• Django
• Docker
• Python
• Sentence Embeddings (Arabic)