zakariaayl/HackAi_ZHO_logs

AI-driven Darija assistant bridging citizens and bureaucracy built at HackAI Hackathon by UM6P & 1337AI empowering Moroccans with voice & web-powered access to administrative info. ## 🧠 Project Idea This project is a **conversational AI assistant** designed to help Moroccan users easily access information about administrative documents and procedures in their own language, **Moroccan Darija**. ### ❓ Why This Idea? In Morocco, many people struggle to find clear and accessible information about government paperwork — such as what documents are needed for a passport, ID card, or other official procedures. This problem is especially challenging for those who: - Prefer to communicate in Moroccan Darija rather than formal Arabic or French, - Have limited experience using digital platforms, or - Face difficulties navigating complex official websites and bureaucracy. ### ✅ What Does the Project Do? - The assistant accepts **user questions in Moroccan Darija via text or voice input**. - It includes **transcription capabilities** that convert voice inputs in Moroccan Darija into text. - It uses a powerful language model (Google’s Gemini model) combined with a web search tool that fetches current, relevant information from Moroccan websites using SerpAPI. - It processes and summarizes search results to provide clear, concise answers that are easy to understand. - It maintains conversation history to offer context-aware responses, supporting follow-up questions and clarifications. ### How It Works (Flowchart Description) User Input: The user submits a question or voice input in Moroccan Darija. Voice Transcription (optional): If the input is voice, it is transcribed into text. Query Processing: The text input is sent to the language model agent. Search Tool Invocation: The agent checks if a web search is necessary and uses the custom SearchWebTool to query relevant Moroccan websites through SerpAPI. Information Retrieval: The search tool fetches and summarizes relevant documents and procedures. Response Generation: The language model composes a clear and context-aware response based on search results and conversation history. User Output …