AI Research Application (Healthcare)

A healthcare consulting company needed a systematic solution for researching and evaluating medical and operational equipment. Previously, information was manually gathered from internal documentation and public sources – a time-consuming process with inconsistent results.

The Solution

The developed application is an AI research chatbot specifically designed for this use case. The chat accesses various data sources: internal documentation, public databases and web searches. The collected information is classified according to defined criteria and this classification is structurally justified.

Architecture

The architecture is based on Retrieval Augmented Generation (RAG) with Semantic Search. Internal documents were embedded and stored in a Milvus Vector Database. For web search, a search engine specifically optimized for this use case was developed: search results are collected in real-time, embedded in-memory and semantically searched. This enables token-optimized chunks for the AI, which generates well-founded research answers based on actually relevant information.

Technical Implementation

The frontend offers a chat interface similar to ChatGPT. The FastAPI backend orchestrates the LangChain-based agent, which selects the appropriate tools depending on the query: Semantic Search on internal documents, web search with subsequent in-memory indexing or database queries.