> **HeartBridge | Case Study | teamvince** > Markdown representation of https://teamvince.com/heartbridge > HeartBridge: a voice-driven cardiac rehab companion, built in a day with a team of strangers, that won the Pulse Foundry AI NYC Healthcare hackathon. Case Study # HeartBridge A voice-driven cardiac rehab companion built to close the gap between clinical eligibility and real-world follow-through. Hackathon Winner Judged build: Lovable · Factory AI · Amp · browser voice ## What this proves - **Workflow:** Cardiac rehabilitation: patients on a recovery program need consistent guidance and check-ins from their care team. - **Build judgment:** Voice-first interface for a population that struggles with screens. Scoped to a single check-in flow before adding scheduling or telemetry. - **Technical pattern:** A voice-first interaction prototype with mock data, plus a clearly separated follow-on architecture for model reasoning and higher-quality voice. - **Buyer relevance:** If you're building for users who can't or won't use a screen-first interface (older patients, accessibility cases, hands-busy workflows), voice-first AI agents need different scoping than chat. This is the shape. ## Submission state and later direction **Judged build, March 29:** The demo used mock, hardcoded data, no backend, and the browser's `SpeechSynthesis` voice. It still won the Pulse Foundry AI NYC Healthcare hackathon. Those are the limitations recorded in the [frozen Devpost submission](https://devpost.com/software/heartridge). **Follow-on direction:** The Claude and ElevenLabs sections below describe the architecture explored after or proposed beyond that frozen submission. They aren't evidence that those services were integrated in the judged build. ## The Problem Cardiac rehabilitation can improve outcomes after a heart event, but participation is low. An [American Heart Association study](https://newsroom.heart.org/news/only-one-in-four-medicare-patients-participate-in-cardiac-rehabilitation) found that roughly one in four eligible older adults with Original Medicare participated in outpatient cardiac rehab. That result has a specific population and study period; it isn't a universal 75% rate. The participation gap has multiple causes, including health status, referral and program availability, transportation, cost, time, confidence, and support between visits. HeartBridge explored one part of that problem: making check-ins and progress more accessible outside a facility. ## The Insight Information alone doesn't remove every participation barrier. A warmer, voice-driven companion could make check-ins easier, coach people through assigned exercises, and keep an approved care circle informed between appointments. It wouldn't replace clinical care or solve access on its own. ## The Solution HeartBridge is a prototype concept that combines voice coaching, daily progress tracking, and care-circle support in one recovery experience. The judged frontend demonstrated that flow with mock data and browser speech. The follow-on product direction adds model-assisted check-ins and a higher-quality voice layer, with clinical review and consent controls required before real patient use. ## How It Works 1. Step 01 ### Voice-guided sessions The judged prototype used browser speech to walk through the coaching flow. ElevenLabs is the follow-on voice option, not a submission-day integration. 2. Step 02 ### Daily check-ins The prototype showed short conversational check-ins with mock responses. A real version would need a reviewed data model, consent, and escalation rules. 3. Step 03 ### Progress visibility The frontend showed recovery milestones, session history, and trends using hardcoded demo data. 4. Step 04 ### Care circle The prototype demonstrated a care-circle view. Real sharing would require explicit patient consent, identity controls, and a secure backend. ## Judged architecture and follow-on path ### Voice Layer The judged build used browser `SpeechSynthesis`. ElevenLabs is the follow-on option for a more natural voice experience. ### AI Backbone Claude is the follow-on reasoning option for interpreting check-ins and adapting tone. The judged submission had no backend or live model integration. ### Frontend Built with Lovable for rapid prototyping, then re-architected in Factory AI using Opus for a solid, scalable foundation. ### Team Coordination Amp smart mode handled PR reviews and merges from 5 teammates, keeping the codebase coherent across a team of strangers. ## Build Process HeartBridge was built in a single day at the Pulse Foundry AI NYC Healthcare hackathon on March 29, 2026. I was placed on a randomly assigned team of 6 strangers. I designed roughly 80% of the product, led the overall build, and presented the winning demo. The build started with a Lovable MVP to get the UI shell up fast. I then combined that repo with health-domain work from a prior project (VitalMed) in Factory AI using Opus 4.6, re-architecting from a README implementation plan. During the build, I reviewed and merged 5 teammate PRs using Amp smart mode to keep the codebase consistent. The result was a frontend prototype with browser voice, progress views, and a care-circle dashboard, all presented live to the judges with mock, hardcoded data and no backend. The frozen Devpost record documents those boundaries. ## Key Design Decisions - **Voice-first over text-first:** Cardiac rehab patients skew older and are recovering physically. A voice interface is lower friction than typing and feels more like a real coach checking in. - **Care circle over solo tracking:** Recovery doesn't happen in isolation. Giving family members and clinicians visibility creates a support network without burdening the patient with manual updates. - **Warmth over clinical precision:** The tone is encouraging, not medical. HeartBridge isn't replacing a cardiologist. It's filling the motivational gap between appointments. - **Lovable for speed, Factory for structure:** Starting with Lovable got a visual prototype up in hours. Moving to Factory AI for the re-architecture let us ship something structurally sound by demo time. - **Amp for team coordination:** With 5 strangers submitting code simultaneously, using Amp smart mode for PR review and merging kept the repo from turning into chaos. ## Tech Stack **Show the tools used on this build +** Lovable Factory AI Amp Claude Opus (follow-on) ElevenLabs (follow-on) Browser SpeechSynthesis (judged demo) React TypeScript Vite ## Outcomes - **1st place** at the Pulse Foundry AI NYC Healthcare hackathon. ## What It Demonstrates - **Hackathon leadership:** Designed 80% of the product and led the build with a team of strangers under extreme time pressure. - **Multi-tool orchestration:** Combined Lovable, Factory AI, and Amp into a coherent one-day build pipeline, then documented the Claude and ElevenLabs follow-on path separately. - **Healthcare product judgment:** The product was shaped around a real participation gap backed by clinical data, not a generic "AI for health" concept. - **Team coordination under pressure:** Managed 5 teammate PRs through Amp smart mode while leading the core architecture. Project navigation ## Browse more case studies Previous project **Commons Copilot** Hackathon-room coordination layer with five agents sharing one Intent Space. 2nd place at Betaworks. Next project **WalkRide** New York route planner mixing walking, Citi Bike, and subway into one ranked trip. Apply this pattern ### Need a voice-first AI build for users who can't lean on a screen? **Maps to:** Workflow Prototype Sprint (1–2 weeks) or Agent & Skills Build (2–4 weeks). HeartBridge shows how to build voice-first AI workflows for users who need accessibility-first design. If your product serves a population that needs more than a chat box, or you're prototyping a voice agent for a care team, an ops floor, or a hands-busy workflow, I can help scope and ship the first version. [Scope a voice-first build](https://teamvince.com/consult?source=heartbridge&service=prototype) [View on GitHub](https://github.com/VJDiPaola/heartbridge-garden) [Review my voice-agent scope →](https://teamvince.com/coaching?source=heartbridge) --- Markdown version of [https://teamvince.com/heartbridge](https://teamvince.com/heartbridge). Site context: [/llms.txt](https://teamvince.com/llms.txt) · [/agent-instructions.md](https://teamvince.com/agent-instructions.md) · [/sitemap.xml](https://teamvince.com/sitemap.xml).