AI / Backend Lead
I built the brain and the voice of HearU, end to end. On the AI side, I designed the persona system that turns each loved one (their name, what they call you, how they talk) into a Claude Opus 5.5 system prompt. I wrote the safety rules into every persona: honesty when asked if it’s really them, no invented memories, 988 in a crisis, and nudges back to the real person. I added camera-still vision and local-time awareness so a 2 AM call feels like 2 AM, and I tuned reasoning effort to cut response time from about 6 seconds to about 3.5.
On the voice side, I integrated Cartesia Sonic 3.6 with a per-person speaking pace, and built the voice-cloning endpoint with server-side consent and length checks. I also set up the backend as Python serverless functions on Vercel so the API keys never touch the browser, and I own the repo and the live deployment.
Log in or sign up for Devpost to join the conversation.