Our team built an AI chatbot that converts plain-language questions into accurate database queries. Pharmacy staff check availability, compare prices, view price history, explore pack sizes, and retrieve identifiers like PZN without writing SQL or using technical tools. The system responds instantly and simplifies access to critical product information.
The system captures data from a replicated PostgreSQL database used for safe proof-of-concept development. To support semantic search, product information is also vectorized, enabling accurate, similarity-based queries across key tables.
We implemented a multi-agent architecture that assigns each user query to the appropriate workflow. For data requests, a language model-based SQL generator writes and validates safe, relevant queries, adhering to strict formatting and security rules to ensure data integrity.
Our team built a demo interface that allows pharmacy staff to interact with the chatbot through a simple frontend. It includes quick-access prompts for common questions and displays the generated SQL for transparency. The demo runs securely through the client’s DNS proxy.
