HackStack

Project Info

ReHeart

Devpost

This project did not submit a demo video on Devpost.

Inspiration

Every year, nearly 1 in 4 heart attack survivors is readmitted within 30 days — often due to missed warning signs, medication errors, or delayed follow-up. The most dangerous period isn’t during hospitalization — it’s the silent gap after discharge. We were inspired by a simple question: What if every heart attack survivor had an AI copilot watching over their recovery at home? ReHeart exists to close that dangerous 30-day gap.

What it does

ReHeart is an AI-powered cardiac recovery copilot that monitors patients in the critical 30 days after heart attack discharge. Patients complete daily voice check-ins describing how they feel. ReHeart: Extracts cardiac-relevant symptoms (chest pain, dyspnea, edema, palpitations) Applies structured post-MI risk stratification Maps symptoms to potential clinical deterioration Generates guideline-aligned recommended actions Escalates RED or YELLOW cases to a care team dashboard Instead of reactive care, ReHeart provides continuous, AI-driven post-discharge monitoring.

How we built it

We built ReHeart as a multi-layer clinical reasoning pipeline: Step 1 - Voice Input Web-based voice recording Transcription using Whisper Step 2 - Clinical Intelligence MedGemma deployed on Modal Structured symptom extraction into strict JSON Risk stratification (GREEN / YELLOW / RED) Condition inference (e.g., recurrent ischemia, heart failure decompensation) Guideline-aligned action recommendation Step 3 - Safety Guardrails Conservative escalation rules Higher-risk default when uncertain Human-in-the-loop alerting Step 4 - Care Team Dashboard Patient list with risk badges Alert queue Structured clinical summaries Trend visualization The system is designed as an executable clinical decision engine.

Challenges we ran into

1. Reliable clinical reasoning Getting structured, consistent medical JSON output required tight schema constraints and prompt tuning. 2. Balancing safety vs over-alerting Too sensitive → constant RED flags. Too lax → unsafe system. We had to encode conservative but practical escalation logic. 3. Demo fragility Voice AI, multimodal inputs, and real-time reasoning introduce many failure points. We built fallback pathways and seeded demo data to ensure reliability.

Accomplishments we're proud of

Built an end-to-end AI clinical reasoning pipeline in 18 hours. Transformed cardiology guidelines into executable decision logic. Created structured escalation outputs instead of vague chatbot responses. Demonstrated a realistic Day 5 crisis detection scenario. Designed the system to prioritize safety and human oversight. Most importantly, we built something that feels like it could truly prevent a readmission.

What's next

Short-term: Add longitudinal recovery modeling Improve condition inference confidence scoring Pilot with cardiology practices Mid-term: Expand to heart failure, post-CABG surgery, and arrhythmias Integrate with EHR systems Add real-time escalation capabilities Long-term vision: ReHeart becomes the AI copilot for every cardiac patient’s recovery journey — continuously monitoring, guiding, and preventing avoidable deterioration at scale. Our goal is simple - make surviving a heart attack safer than ever before.

Analysis

Compare with all teams

View

Metric

Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.

Technology

Found in codeClaimed only
  • CSSIn code
  • FastAPIIn code
  • HTMLIn code
  • Next.jsIn code
  • OpenAIIn code
  • PythonIn code
  • ReactIn code
  • SQLIn code
  • SupabaseIn code
  • Tailwind CSSIn code
  • TypeScriptIn code

11 of 11 appear in the indexed code.

AI coding agents

  • Claude CodeConfig

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

126 KB

Source files

18

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.

0 stars