Inspiration

Cancer patients are highly vulnerable to sudden deterioration (like sepsis) when recovering at home between hospital visits. We were inspired to bridge this dangerous monitoring gap by combining continuous wearable data with patient-reported symptoms to catch early warning signs before they become emergencies.

What it does

Our project (Oncology Watch) is a dual-view dashboard. Patients easily log their symptoms, while nurses see a real-time timeline of those symptoms alongside wearable vitals. It acts as a digital triage nurse, calculating clinical risk scores and instantly flagging patients who need urgent care, while filtering out stable patients to reduce alarm fatigue.

How we built it

We used Python and Streamlit to build the prototype. Instead of a traditional database, we engineered a custom timeline simulator that let us safely "fast-forward" through a synthetic 28-day patient journey. The core logic relies on the validated NEWS2-like medical scoring framework, paired with strict data quality gates for wearables, calculated as:

$$\text{Coverage} = \left( \frac{\text{Active Minutes}}{\text{Total Window Minutes}} \right) \times 100$$

(Note: In accordance with hackathon rules, we utilized ChatGPT and Gemini for code generation, documentation, and debugging assistance).

Challenges we ran into

Managing medical timelines across different time zones caused several system crashes until we locked the entire architecture into strict UTC time. We also had to figure out how to safely handle wearable disconnects without relying on AI to "guess" missing health data, ultimately building logic that explicitly rejects poor connections to avoid false alarms.

Accomplishments that we're proud of

We built a rock-solid, 100% passing automated test suite in just 24 hours. We are also incredibly proud to have built a rich, functional prototype using a 100% synthetic dataset, ensuring zero risk to real patient privacy.

What we learned

We learned that in healthcare technology, safety and predictability matter more than complex AI. We realized the immense value of relying on deterministic, universally validated medical rules (like NEWS2) rather than "black-box" machine learning models that clinicians cannot easily audit or trust.

What's next for Oncology Watch

To scale into a production-grade medical device, Oncology Watch will transition to a distributed, HIPAA-compliant cloud architecture. Future clinical wearables will utilize edge computing to process vital streams locally and transmit compressed summaries. The AI will adopt a hybrid model, combining personalized Time-Series Transformers for baseline forecasting and medical NLP for symptom parsing, backed by deterministic NEWS2-like safety guardrails. Backends will leverage Apache Kafka for streaming, PostgreSQL/TimescaleDB for storage, and SMART on FHIR for Epic/Cerner integration. Finally, we will ensure absolute auditability through federated learning and strict version-controlled, non-evolutionary SaMD deployment models.