ClinicalMatch AI

An autonomous AI agent powered by CockroachDB (AWS)

Comment

## Inspiration

Recruiting eligible candidates for clinical trials is one of the biggest bottlenecks in medical research. Patients with complex diagnoses often struggle to find relevant trials, while trial criteria change continuously. Traditional search portals lack "memory"—requiring patients to repeatedly re-enter complex biomarker and medical data without proactive updates when suitable trials launch.

What it does

ClinicalMatch AI is an intelligent agentic workspace that bridges this gap. It maintains a secure, persistent memory of patient profiles, lab values, and biomarkers to continuously evaluate eligibility against open clinical trials. When a new trial matches a patient's historical profile, the agent proactively flags the match and updates its interaction history.

How we built it

  • Database & Agentic Memory: CockroachDB Serverless (on AWS) handles multi-region persistence, storing structured patient profiles, trial criteria, and the agent's contextual interaction logs (agent_memory_log).
  • Cloud Infrastructure: Multi-region deployment hosted on AWS (eu-central-1) ensuring zero-downtime persistence and high availability.
  • Frontend & Workspace: Developed using a modern React/Next.js dashboard interface designed for clinical trial coordinators and patients, featuring real-time biomarker tracking (EGFR, PD-L1, ALK) and trial eligibility scoring.
  • Agentic Backend: Python-based PostgreSQL integration query layer interacting directly with CockroachDB to fetch contextual memory and calculate inclusion/exclusion criteria compatibility.

Challenges we ran into

Designing a memory architecture that balances strict clinical data privacy with flexible query capabilities. Leveraging CockroachDB's PostgreSQL compatibility allowed us to seamlessly query JSON/Array fields representing complex oncology biomarkers without losing performance.

Accomplishments that we're proud of

  • Built a sleek, intuitive workspace that translates complex genomic/lab markers into clear eligibility insights.
  • Successfully implemented persistent "Agentic Memory" via CockroachDB on AWS.

What's next for ClinicalMatch AI

  • Integrating real-time vector search capabilities natively within CockroachDB for semantic matching of raw clinical protocol PDFs.
  • Expanding notification triggers for patients when phase-III trials expand to nearby geographic locations.

Built With

Share this project:

Updates