AskBiotact
AI health consultant for BIOTACT Deutschland. RAG pipeline with 37 documents, async ExtractionAgent, multi-provider LLM and structured order flow.
AskBiotact is an AI health consultant built into the biotact-core-v2 platform. It advises customers on BIOTACT Deutschland products (13 supplements, 7 appliances, 9 accessories) via Telegram bot and Public API.
Each request goes through a full pipeline: authentication, Redis chat history, CRM profile from PostgreSQL, order detection, query enrichment (ExtractionAgent insights, history context, semantic triggers for prices), embedding via text-embedding-3-large, Qdrant vector search (37 documents, threshold 0.30), and GPT-5 mini response generation with CRM context in the system prompt.
After each response, an async ExtractionAgent (GPT-4o-mini, temperature=0) extracts mentioned products, symptoms, family info, intent, and summary — stored in conversation_insights for future query enrichment.
When a phone number is detected, the Structured Order Flow activates: LLM parses name, phone, address, and products → formatted order with prices sent to a Telegram sales group.
