Abdelrahman590/Traffic_AI_Assistant

This is a chat bot customized on traffic rules in Egypt using Langchain and RAG # 🚀 Tips Hindawi Challenge (June–July) 2026 > 🏆 This repository is my official submission for the **Tips Hindawi** **Challenge (June–July) 2026**. ## 👤 Participant | Field | Value | | ---------------- | ------------------------------------ | | Full Name | Abdelrahman Mahmoud Mossad | | Project Name | Intelligent Traffic Law Assistant | | GitHub Username | Abdelrahman590 | | Challenge Batch | June–July 2026 | | Training Program | Large Language Models (LLMs) Program | | Organization | **Edrak for Ai** | --- # 📖 Project Overview The **Intelligent Traffic Law Assistant** is a RAG-based (Retrieval-Augmented Generation) chatbot that answers questions about Egyptian traffic law, including driving license requirements and traffic violations, in both Arabic and English. The project was built entirely on free resources, using open-source models running on Kaggle GPU sessions and exposed through ngrok tunnels, with no local GPU or paid API required. --- # ✨ Features * Answers questions about Egyptian traffic law (driving license requirements and traffic violations) in both Arabic and English * Retrieval-Augmented Generation (RAG) pipeline with source citation display * Two-stage chain: a Prompt Optimizer chain followed by a Retriever + QA chain * Runs entirely on free resources (Kaggle GPU + ngrok), with no paid API or local GPU needed * Streamlit-based frontend with an "Optimize Prompt" button --- # 🛠️ Technologies Used * **Compute:** Kaggle GPU sessions (free tier) + ngrok tunnels * **LLM:** Qwen2.5-3B-Instruct * **Embeddings:** bge-small-en-v1.5 * **Backend:** FastAPI + pyngrok (served from the Kaggle notebook) * **Orchestration:** LangChain , with custom `KaggleRemoteLLM` and `KaggleRemoteEmbeddings` wrappers routing inference via HTTP to the Kaggle ngrok URL * **Vect …