Cohort Lens

Inspiration

Finding the right participants for public health and clinical studies is often a slow, manual, and inefficient process. When I was working in a public health study, a significant amount of time was spent:

• Manually identifying and screening candidates who met specific criteria (demographics, medical history, etc.)
• Contacting past study participants who had expressed interest in participating again

Imagine an AI-powered solution—one that could autonomously explore medical datasets, uncover hidden connections, and recommend ideal study participants.

What it does

Cohort Lens is an agentic AI application that streamlines clinical and public health study recruitment using GraphRAG:

• Graph Intelligence (NVIDIA cuGraph, ArangoDB): Maps complex relationships between patients, conditions, and treatments to uncover hidden insights.
  
• LLM-Powered Reasoning (LangChain, LangGraph, OpenAI): Dynamically retrieves relevant patient data and applies intelligent filtering based on study criteria.
  
• Agentic Workflow: Cohort Lens autonomously explores the data, refines queries, and optimizes cohort selection—minimizing manual effort.
  

How it was built

• Data Processing: Synthea's synthetic medical data was used to simulate real-world patient records.
  
• Graph Construction: Built a knowledge graph in ArangoDB, using NVIDIA cuGraph for scalable graph analytics.
  
• GraphRAG Implementation: Combined graph-based retrieval with LLM-powered reasoning using LangChain and LangGraph.
  
• Agentic AI Logic: Designed an autonomous agent workflow that iterates over patient data to continuously refine cohort selection.
  

Challenges

• Efficient Search & Retrieval: ArangoSearch was crucial for efficiently searching for relevant nodes via the description field, offering full-text search capabilities that made it easier to locate specific conditions, treatments, and patient data.
  
• Building Agentic Workflows: Ensuring that LangGraph-powered agents retrieved relevant and contextually appropriate study participants.
  

What's next for Cohort Lens

• Integrating Real-World Healthcare Data: Move beyond synthetic datasets and test on FHIR-compatible real-world data.
  
• Enhancing Agentic Capabilities: Allow the AI to autonomously suggest and refine study criteria based on past trial success.
  

Cohort Lens transforms study recruitment into an intelligent, automated, and AI-driven process—accelerating medical research and unlocking new possibilities for precision medicine.

Built With

• arangodb
• langchain
• langgraph
• python