Inspiration
Stress and cravings do not just appear. Your body shows them first, through increased sweating, a racing heart, and falling heart rate variability. Nerva utilizes this fact to prevent stress before it becomes an irreversible cycle. First, it catches the stress early, helps you calm down in the moment and rewards you for recovering, because positive reinforcement actually works.
What it does
What it does
Nerva is an AI wristband (simulated for this hackathon) that detects stress-related arousal and pays a real reward when the wearer recovers.
Phase 1 (always on, low power): a small neural network watches sweat, sweat bursts, skin temperature and motion. If Phase 1 sees early signs, it switches on the heart sensor. Phase 2 adds heart rate and heart-rate variability to confirm the stress. The band buzzes and starts guided breathing. If the wearer stays calm for 60 seconds, the band signs a claim, and a Solana program pays a parent-set reward in a USD stablecoin from the parent’s pool straight to the child’s wallet.
How we built it
Nerva is a multifaceted AI model that was trained on several online datasets. The synthesized model contains two smaller models that are gated based on the stress phase. During the first phase, the AI tracks low-power hardware that is computationally light. Once the primary sensors confirm the possibility of stress, the second gate activates to consolidate the decision of stress. This allows the device to have more longevity per charge by limiting the use of computationally heavy devices. To positively reinforce management of stress, we also utilized Solana, an Anchor Program with a PDA-owned reward pool and an SPL token vault that only pays when an individual successfully defeats a wave of stress.
Challenges we ran into
- Lack of accuracy
- Finding a model that can be quantized to 8 bits and fit in the device.
- Finding a way to reward the user for their amelioration of stress
Accomplishments that we're proud of
- A full working loop with contact between the AI, wearer, and Solana
- The program, not a server, controls the rewards ## What we learned Lab accuracy does not always hold up in real life ## What's next for Nerva
- Build the hardware version of the Nerva Band or Nerva Band v1 on an ESP32 with sweat, temperature, motion, hear sensors, power gating, and a vibration motor, for under $30
- Improve the accuracy of the model
- Create an interactive integrated app
- Spread this into my community by partnering with families, schools for exam stress, and clinics for recovery programs
Built With
- anchor
- css
- esp32
- flask
- github
- gunicorn
- html
- javascript
- json
- lite
- litert
- numpy
- python
- render
- rust
- scikit-learn
- solana
- solders
- spl
- tensorflow
- tinyml
