Unshaken – Parkinson’s Monitor

Inspiration

Parkinson’s disease is one of the fastest-growing neurological disorders worldwide, affecting tens of millions and steadily taking away independence. Patients told us about tremors, freezing episodes, and the fear of being alone when something goes wrong. We wanted to build a solution that brings back safety, dignity, and freedom.

What it does

Unshaken transforms an everyday smartwatch into an intelligent Parkinson’s companion that continuously monitors tremors, detects freeze-of-gait in real time, and responds instantly when the patient needs help. The system can trigger an automated check-up call through an AI agent, notify guardians and caregivers, and generate meaningful reports for doctors. It also provides rhythmic auditory stimulation through a built-in metronome to help patients break out of freezing episodes. Everything works together to safeguard patients throughout their day without requiring extra devices.

How we built it

We designed a two-layer architecture that combines on-device sensor intelligence with cloud-based AI agents. On the smartwatch, we processed gyroscope, accelerometer, and heart-rate streams to approximate a Freeze Index and identify tremor patterns while maintaining battery efficiency. The watch was developed using HarmonyOS and Oniro, taking advantage of their sensor APIs and low-latency audio features.

In the cloud, we built a coordinated set of agents responsible for interpreting symptoms, performing check-up calls, deciding whether escalation is needed, and creating reports for doctors and family members. A lightweight dashboard visualizes tremor trends, freeze events, medication adherence, and stress-level patterns.

Challenges we ran into

Building accurate freeze-of-gait detection with noisy motion data was one of the hardest tasks, especially while keeping the algorithms fast enough to run continuously on a watch. We also had to balance real-time responsiveness with battery constraints, and we spent significant time tuning the interaction between multiple AI agents so they behave consistently and safely. Designing an interface simple enough for elderly patients, yet informative enough for clinicians, was another key challenge.

Accomplishments that we're proud of

We successfully built a working end-to-end prototype that runs on a consumer smartwatch and reliably detects tremors and freeze-of-gait in real time. We implemented a fully functional AI emergency call flow that checks on the patient when unusual patterns appear. The multi-agent architecture proved effective in turning raw motion data into actionable medical insights. We also developed a dashboard that clinicians and families can use to track long-term progression.

What we learned

We learned how powerful wearable sensors can be for understanding neurological symptoms when combined with intelligent analysis. We realized that Parkinson’s varies significantly from person to person, and adaptive models are essential. We also saw how AI agents can meaningfully reduce caregiver workload while increasing patient safety. Finally, we validated how rhythmic auditory stimulation can genuinely help break freeze episodes in real-life scenarios.

What's next for TUMCooked

Our next steps include clinical collaboration with neurologists and rehabilitation centers to validate the system at scale. We plan to introduce personalized models tuned to each patient’s movement patterns, expand the system with fall-risk scoring and rehabilitation exercises, and optimize on-device ML to reduce latency even further. Ultimately, we aim to evolve Unshaken into a medical-grade monitoring tool that patients and clinicians can rely on every day.