ClinicalTrial MatchMaker

Inspiration# Clinical Matchmaker — Project Story

About the Project

Clinical Matchmaker was inspired by a simple but painful reality: patients often miss life-saving clinical trials not because they are ineligible, but because eligibility criteria are locked inside dense, unstructured documents. ClinicalTrials.gov contains thousands of trials, yet the burden of interpreting inclusion and exclusion criteria falls on patients and clinicians who already face time, cognitive, and emotional overload.

I wanted to explore a different question:
What if AI could reason over clinical trial criteria the way a careful human advocate would—step by step, transparently, and compassionately?

That question became the foundation of Clinical Matchmaker.

What I Learned

This project fundamentally changed how I think about building AI systems for high-stakes domains like healthcare.

I learned that:

Reasoning clarity matters more than raw accuracy when outcomes affect human lives.
Treating prompts as first-class artifacts (via Prompt-Driven Development) makes AI systems more explainable, testable, and adaptable.
Multi-agent systems work best when each agent has a narrow, well-defined responsibility and communicates through strict schemas.
Empathy is a design constraint, not a UI polish step—especially when delivering medical information.

How the Project Was Built

Clinical Matchmaker is built using Prompt-Driven Development (PDD), where prompts are the source of truth rather than code.

The system is composed of four specialized agents:

A Trial Scout agent that discovers actively recruiting trials.
An Eligibility Extractor agent that parses inclusion and exclusion criteria.
A Clinical Matching Specialist that evaluates eligibility using transparent scoring.
A Patient Advocate agent that translates results into plain language and voice.

Each agent is defined by a .prompt file specifying its role, inputs, outputs, constraints, and examples. From these prompts, code and tests are generated and regenerated as the system evolves. Toolhouse orchestrates agent execution, rtrvr.ai handles structured web extraction, and ElevenLabs produces compassionate voice explanations.

Challenges Faced

The hardest challenge was avoiding false certainty. Clinical eligibility is rarely binary, so the system had to explicitly model uncertainty instead of hiding it. Another major challenge was ensuring that explanations remained helpful without crossing into medical advice.

Balancing technical ambition with ethical responsibility was the most important—and most rewarding—part of this project.

Closing Reflection

Clinical Matchmaker is not about replacing clinicians. It is about lowering the barrier to discovery, giving patients clearer paths forward, and demonstrating how agentic AI—when designed with structure, humility, and care—can make complex systems more humane.