An intelligent Tunisian law expert and compliance monitoring system powered by LLMs and a curated Tunisian legal dataset.
It features interactive contract analysis using OCR and RAG-based document understanding.

Combot was developed as part of a university project for the Contract Lifecycle Management module within the Management Information Systems program at Esprit School of Engineering.
It leverages modern technologies to automate the analysis, monitoring, and compliance enforcement of legal documents--reducing human error and operational risks.

This project aims to make information about Tunisian laws related to contract compliance more accessible and interpretable, with the capability of updating autonomously.

• Uses Langgraphe to create a robust, non-sequential workflow
• Extracts relevant references from Tunisian law to answer contract-related questions
• Crawls the web to detect updates in Tunisian laws and checks for relevance to the targeted legal codes
• Automatic extraction of contractual obligations using OCR and NLP
• Semantic document search using Qdrant and hybrid search techniques
• Response generation via LLMs using Retrieval-Augmented Generation (RAG)
• Lightweight backend implemented in Flask

• Python
• Flask
• Qdrant (Vector Database)
• Langgraphe, PyPDF2, and other RAG pipeline tools
• llama, DeepSeek, BAAI/bge-m3, BM25 for model and search components

• OCR (PaddleOCR) for scanned document processing
• NLP for understanding contract clauses
• LLMs for chatbot interactions and legal response generation
• Hybrid Search for both dense and keyword-based retrieval
• RAG (Retrieval-Augmented Generation) to generate answers with contextual document references

• Dataset of contracts (from Esprit University)
• Tunisian legal codes, including:
  • loi_defense_contre_pratiques_deloyales_importation
  • loi_relative_Startups
  • loi_relative_commerce_exterieur
  • loi_societes_commerce_international
  • loi_societes_ligne
  • texte_code_societes_commerciales

/ bm25_states
/ data
/ evaluation
/ index
/ legal_codes
/ log
/ static
/ css
/ stores
/ templates
README.md
app.py
combot_functions.py
config.py
contract_functions.py
db.py
webCrawlerAgent.py
requirements.txt

1. Add your Groq API key in config.py
2. Install requirements:
   pip install -r requirements.txt
   
3. Start Qdrant using Docker:
   docker run -p 6333:6333 -p 6334:6334 -v "${PWD}/qdrant_storage:/qdrant/storage" qdrant/qdrant
4. Run the application:
   python app.py
   

Special thanks to Esprit School of Engineering for the opportunity to build this project and for providing access to the DGX platform, which allowed us to experiment with fine-tuning LLMs using legal texts.

We are grateful to:

• Mr. Morad Zerrai and Mrs. Nardin Hannfi for their mentorship and guidance throughout the project

• Mr. Souhail Oueslati, Group Chief Financial Officer, for offering expert legal and financial advice that shaped our work