An intelligent assistant for radio sales teams.
Commbot is a conversational assistant built for the sales team of a radio advertising agency. It centralises business knowledge and streamlines client meeting preparation through a multimodal RAG architecture.
Architecture and ingestion
The system ingests diverse sources — PDF documents, WordPress pages, images and videos — to build a rich, up-to-date knowledge base. Hybrid search combines semantic and full-text approaches, with French-specific linguistic normalisation and multilingual embedding models.
Meeting preparation
The flagship feature lets sales representatives prepare client meetings: the assistant automatically aggregates relevant information about an advertiser — history, industry, tailored offers — to generate a structured, source-backed briefing document.
Technical stack
The product is built with Python and FastAPI for the API layer, Supabase with pgvector for vector storage, the Claude API (Anthropic) for contextualised response generation, and Coolify for self-hosted deployment.