AskLevant The Pragmatic AI Document Assistant for Social Work

At the Fondation du Levant, social workers and administrative staff lose precious time (often 15+ minutes per search) hunting down procedures, regulations, or forms in complex network drives. This lost time directly impacts our core mission of helping beneficiaries. The Challenge: Design AskLevant, an internal chatbot you can query in natural language ("Where is the leave request form?"). Based on a RAG (Retrieval-Augmented Generation) architecture, it searches our internal knowledge base (French documents), generates a precise answer, and provides the exact document link. Technical Goals for the Hackathon: Pragmatism: IT resources in the social sector are limited. The goal is to find the most efficient and maintainable architecture (integrating existing tools, lightweight models). Ingestion: Process various formats (PDF, Word) with minimal manual effort. Zero Hallucination: For strict internal procedures, the AI must never invent rules. Anti-Shadow IT: Act as a writing assistant (spell check, rewording) to prevent sensitive data from being pasted into public AIs. Bonus (Human Fallback): If the AI doesn't know the answer, it auto-generates a pre-filled email or ticket for the right department. The Impact (Open Knowledge): The technical "recipe" developed during these 48h will be documented and shared with other regional foundations so they can deploy their own assistant without reinventing the wheel!

Guest