GutGuru is a clinical insight layer that gives doctors a clearer picture of gut health over time. The engine rests on one honest idea about how sparse personal health data really behaves. Lab panels arrive only every few months, so they anchor against a dense daily symptom series on one shared timeline. The system flags out of range values across every report and clinical finding a patient uploads. It highlights the specific places where two independent reports happen to agree with each other. It also detects when symptoms quietly shifted after a supplement routine or a diet changed. Every output stays firmly on the observation side of the clinical line.
On the technical side I wrote the full pipeline in Python with Notion serving as the structured data layer. Open models served through Nebius narrate the finished findings back in plain readable language. PubMed supplies real citations sitting behind a provider agnostic evidence interface for trust. A caching and lazy compute design keeps inference cost near zero until a clinician asks for an overview. I built this entire thing solo for the Nucleate New York BioHack under the NextGen BioAgents program.
