Inspiration Natalie, a gastroenterology fellow, inspired this project after facing documentation and administrative burdens that slowed her workflow and delayed patient communication. For example, when colonoscopy pathology reports return, clinicians must manually review, interpret, and communicate results—a repetitive process that wastes time and risks delays in care.
What we learned We learned to create agentic clinical workflows using PhenoML for data interpretation and DeepL for multilingual, patient-friendly summarization.
How we built it PhenoML for phenotype extraction and report analysis LangGraph + GPT-5 for agentic reasoning and automation DeepL for multilingual summarization Epic-style UI mock for notifications, care plans, and summaries Automated pipeline from ingestion to reasoning to output
Challenges Properly prompting and tuning the agent to accurately interpret unstructured pathology and lab data Ensuring high factual accuracy and consistency in generated clinical summaries and reports Managing complex data mapping between unstructured text and structured clinical formats Simulating realistic EHR workflows using synthetic Medplum data for testing and validation
Built With
- deepl
- gpt-5
- phenoml
- python
- typescript
Log in or sign up for Devpost to join the conversation.