AI Innovations
AI Innovations
Clinical technology projects built from frontline mental health practice, focused on crisis prevention, care continuity, accessibility, and safety for people who may not be reached by standard app-based care.
Project Lead
Jun 2026 - Present
TheraCat: Proactive AI Wearable
First Place, DDOSKI's Lab Track and SkyDeck Grand Prize, 2026 Cal AI Hackathon; guaranteed Berkeley SkyDeck Pad-13 incubator admission
- Built a two-part panic-intervention prototype: a wrist-worn sensor and a soft plush companion that respond before the user has to unlock a phone or open an app.
- Translated clinical experience with anxiety and panic attacks into a zero-friction support loop: sense agitation, classify rising arousal, and respond through touch, voice, and grounding support.
- Designed haptic paced-breathing support on the wrist plus an AI voice companion that can check in, guide grounding skills, and route risk language toward crisis resources.
- Led clinical logic and user experience, contributed to hardware prototyping, and helped connect the clinical problem, wearable workflow, and pitch narrative.
Clinical logic
- Designed for the moment when panic reduces a person's ability to read instructions, navigate screens, or initiate help-seeking.
- Centered somatic co-regulation: slow haptic breathing cues, physical presence, and a calm companion instead of another app workflow.
- Built in safety thinking from the start, including risk-language screening and direction toward 988 crisis support.
Technical build
- ESP32 and MPU6050 wristband streams accelerometer / IMU motion data through BLE or WebSocket pathways.
- Motion windows are processed into calm-versus-agitated features, including variability and accelerometer-axis patterns.
- Python host and orchestrator components connect sensing, stress scoring, plushie haptic actions, and adaptive breathing prompts.
- Prototype includes notebook-based ML exploration, BLE CSV capture, firmware paths, and a live orchestrator architecture.
Clinical UX
Panic Intervention
Wearables
ESP32
MPU6050
BLE
Python
Haptics
AI Voice
Project Media
Project Lead
Mar 2026 - Present
TheraCaIl: Automated Voice-AI Outreach Platform
First Place, Health Track, HackMerced XI
- Built a voice-AI treatment alliance aid for clinicians whose patients struggle to complete action plans between therapy sessions.
- Designed AI-powered phone check-ins that ask about basic status, depression and anxiety indicators, safety concerns, and action-plan progress.
- Created clinical risk logic so urgent responses, missed calls, and high-risk signals are surfaced for clinician review instead of buried in unstructured notes.
- Served as the clinical authority in a high-intensity software build, translating frontline care-continuity gaps into product logic and dashboard workflows.
Clinical logic
- Built around the clinical reality that therapy action plans often happen hundreds of hours away from the next session.
- Designed for accessibility: patients do not need to download an app, log in, or navigate a screen; they only answer a phone call.
- Supports therapist follow-up by summarizing patient responses, action-plan progress, missed calls, and safety concerns.
Technical build
- Frontend uses React, TypeScript, Vite, Tailwind CSS, and lucide-react.
- Backend uses Python, FastAPI, WebSockets, MongoDB, and webhook-based transcript processing.
- AI summary layer uses Gemini; voice AI uses ElevenLabs; phone integration uses Twilio through ElevenLabs outbound calling.
- Dashboard organizes patient records by risk level and updates with recent AI-generated summaries and warning signals.
Care Continuity
Clinical Safety
Voice AI
ElevenLabs
Gemini
Twilio
FastAPI
MongoDB
React