What Is the MedGemma Impact Challenge?
The MedGemma Impact Challenge called on developers worldwide to build prototype healthcare applications using Google’s open medical AI models. Google launched this initiative through its Health AI Developer Foundations (HAI-DEF) program in late 2024. The program provides open-weight models to help solve real-world clinical challenges at scale.
Furthermore, the challenge built on the success of MedGemma, Google’s most capable open model for Health AI development. In January 2025, Google introduced MedGemma 1.5 and then partnered with Kaggle to launch the challenge. Notably, over 850 teams submitted entries from across the globe. Together, these teams demonstrated remarkable potential to reshape healthcare delivery worldwide.
Grand Prize Winners
The selection committee chose four outstanding teams as grand prize winners. Each solution tackled a distinct, high-impact healthcare problem.
EpiCast — Bridging Disease Surveillance Gaps in West Africa
EpiCast is a mobile-first demo built for the Economic Community of West African States (ECOWAS). It uses a fine-tuned MedGemma model alongside MedSigLIP and HeAR to help community health workers convert unstructured clinical notes in local languages into structured WHO Integrated Disease Surveillance and Response (IDSR) signals. Consequently, this enables the early detection of disease outbreaks in underserved regions.
Sunny — Privacy-First Skin Health Tracker
Sunny is a mobile-first demo that empowers users to self-examine skin changes for potential signs of cancer. Using a fine-tuned MedGemma model, it generates structured reports directly from skin photographs. Additionally, it preserves user privacy throughout the entire process. This makes it both accessible and trustworthy for everyday consumers.
FieldScreen AI — On-Device TB Screening
FieldScreen AI addresses tuberculosis screening in resource-limited settings. It uses a fine-tuned MedGemma model to analyze chest X-rays and a HeAR-based classifier to detect TB signs in cough audio. Moreover, the system runs entirely on-device, using MedASR for voice input and TranslateGemma for local language output. Therefore, it works even in areas with no internet connectivity.
Tracer — Preventing Dangerous Medical Errors
Tracer demonstrates an AI-driven workflow to reduce medical errors in clinical settings. It extracts hypotheses from physician notes and reconciles them against incoming test results. The model then assigns confidence scores to flag discrepancies or incomplete tests for immediate human review. As a result, clinicians can catch critical oversights before they cause patient harm.
Special Technology Winners
These awards recognize innovation across three themes: agentic workflow improvement, fine-tuning for novel tasks, and edge-AI on-device deployment.
ClinicDX — Offline AI for Sub-Saharan Africa
ClinicDX integrates a custom fine-tuned MedGemma model into OpenMRS for health centers in sub-Saharan Africa. Impressively, it runs entirely offline and queries over 160 WHO and MSF guidelines to deliver clinical AI support without internet access.
UniRad3s — Streamlined Radiology Workflows
UniRad3s fine-tunes MedGemma and incorporates MedSAM2 to simplify radiology into three pillars: Spot (anomalies), Segment (3D lesions), and Simplify (patient-friendly reporting). Similarly, it reduces radiologist workload while improving report clarity.
BridgeDX — Offline Decision Support for Emergencies
BridgeDX draws inspiration from the 2015 Nepal earthquake. It anchors its reasoning in WHO, MSF, and Orphanet rare disease guidelines to support health workers and first responders in crisis situations.
CaseTwin — Faster Referrals in Rural Hospitals
CaseTwin matches acute chest X-rays with historical “twin” cases using an agentic workflow. This approach cuts a previously hours-long manual retrieval process down to just a few minutes. Consequently, rural hospitals can accelerate referrals and improve patient outcomes.
BigTB6 — Voice-Driven TB and Anemia Screening
BigTB6 is a voice-driven screening demo for tuberculosis and anemia. It identifies critical biomarkers through cough analysis, chest X-ray review, and physical pallor assessment. Notably, it targets triage support in resource-constrained environments where specialists are scarce.
Honorable Mentions
The selection committee also recognized four additional entries for their ingenuity and innovation.
- Dual Path ICU — a demo designed to manage high-intensity critical care workflows.
- Sentinel — an on-device mental health monitoring demo for veterans between clinical visits.
- Enso Atlas — a clinical decision support demo designed to assist pathology workflows.
- CAP CDSS — an agentic assistant demo for managing Community-Acquired Pneumonia in high-pressure environments.
What’s Next for MedGemma and HAI-DEF
The MedGemma Impact Challenge proved the immense versatility of Google’s open health AI models. These teams built solutions that previously required far more resource-intensive, ground-up development. Moving forward, Google plans to continue spotlighting how developers use HAI-DEF models to bridge healthcare gaps globally. To follow the journey,
