How my students are using AI to solve real problems

Students shouldn't build production systems.

A student in my class just proved me wrong.

One of my grad students benchmarked five large language models for structured data extraction from ClinicalTrials.gov. This wasn't a toy project. Clinical trial evidence synthesis drives treatment decisions for millions of patients.

She tested retrieval accuracy across 200+ trial records. The best models retrieved relevant eligibility criteria that human reviewers missed. Her ensemble approach doubled retrieval rates while maintaining near-perfect inter-model agreement.

The technical stack: Python, LangChain for prompt orchestration, custom evaluation metrics for extraction precision, and systematic comparison of GPT-4, Claude, and smaller open-source alternatives.

She didn't just build a model. She built a reproducible evaluation framework that our research group now uses.

This is happening in UNLV classrooms right now. Students aren't learning theory in isolation. They're deploying systems that healthcare researchers actually need.

If students are building this, imagine what professionals can do with these approaches.

What real-world problems are you seeing students tackle with AI?

#AIResearch #HealthcareAI #UNLV