Case 2 of 4, Live
ClinicalAgent
A multi-agent platform for clinical trial intelligence: patient-to-trial matching, FDA safety analysis and streaming answers to clinical questions.
Overview
ClinicalAgent helps healthcare professionals reason over clinical trial data and drug safety information. Four specialised agents (enrollment, safety, efficacy and general) work under LangGraph orchestration to answer complex clinical queries, match patients to trials and surface FDA adverse-event patterns.
Python / FastAPI / LangGraph / Next.js / MongoDB / Neo4j / ChromaDB / Groq API
Problems, and what I built
- 01
Routing ambiguous clinical questions to the right specialist, reliably
An intent-classification layer that dispatches before the agent graph runs
- 02
Combining vector, graph and document retrieval without latency spikes
Parallel retrieval across ChromaDB, Neo4j AuraDB and FAISS, with merged results
- 03
Streaming partial answers while keeping every session isolated
A FastAPI server-sent-events endpoint with per-user sessions in MongoDB
- 04
Predicting trial-match success with little labelled outcome data
A model trained on ClinicalTrials.gov outcomes and openFDA signal patterns
What it does
- LangGraph orchestration of four specialised agents
- Patient-to-trial ML matching with success-rate prediction
- Streaming responses with live agent visualisation
- Vector search, graph analytics and a document store working together
- JWT-secured sessions with full conversation history
Result
Hours of manual trial search and safety review become a conversation, which shortens the path from question to insight for clinical researchers.