HeartNote

AI-powered postpartum health platform for faster symptom recognition and clearer doctor communication.
Natural language symptom input translated into structured clinical insights and cardiovascular risk analysis.
Guided postpartum intake flow collecting symptom history and cardiovascular risk factors.
Interactive body map for logging symptom location, severity, and cardiovascular warning signs.
Optional face scan detecting visible indicators like swelling and pallor using MediaPipe Face Mesh.
Patient-friendly report translating clinical findings into clear, understandable postpartum health insights.
Doctor-facing SOAP report formatted for faster clinical review and clearer postpartum risk communication.

Inspiration

Many postpartum women with cardiovascular complications are dismissed or misdiagnosed because symptoms like chest pain, swelling, dizziness, or shortness of breath are often seen as “normal” after childbirth. Delays in diagnosis can become life-threatening, especially for conditions like postpartum preeclampsia or peripartum cardiomyopathy. We wanted to build a tool that helps mothers communicate symptoms more clearly and receive earlier medical attention.

What it does

HeartNote is an AI-powered postpartum health website that helps new mothers track symptoms and identify potential cardiovascular warning signs. Users can describe symptoms via natural-language input, interact with a body map, and use a camera-based face scan to detect visible indicators such as swelling or pallor. HeartNote then generates AI-powered insights, cardiovascular risk flags, and doctor-ready SOAP summaries that make communication with healthcare providers faster and clearer.

How we built it

We built HeartNote using React and Next.js for the frontend and Next.js API routes for the backend. We integrated the Claude API to power symptom interpretation, clinical translation, risk analysis, and SOAP report generation. For computer vision, we used MediaPipe Face Mesh and TensorFlow.js to detect visible facial indicators directly in the browser. We also implemented an interactive body map, PDF export, and a structured AI pipeline to generate both patient-friendly and doctor-facing reports.

Challenges we ran into

One of our biggest challenges was balancing medical accuracy with accessibility. We had to carefully design prompts and risk logic that translated everyday language into clinically meaningful information without making unsupported medical claims. Another challenge was integrating multiple input systems, including voice, body map interactions, AI processing, and face scan analysis, into a smooth real-time workflow within a short hackathon timeline.

Accomplishments that we're proud of

We are proud that HeartNote goes beyond being a generic symptom checker. The platform combines AI symptom translation, cardiovascular risk analysis, interactive body mapping, and visual face scan indicators into a single workflow designed specifically for postpartum care. We successfully built a system that generates realistic doctor-style SOAP reports while still remaining understandable for patients. Most importantly, we created a tool focused on helping women feel heard and taken seriously.

What we learned

We learned how difficult healthcare communication can be, especially in postpartum care where symptoms are frequently overlooked or normalized. We also gained hands-on experience integrating AI into a real-world healthcare application while thinking carefully about ethics, usability, accessibility, and clinical responsibility.

What's next for HeartNote

Next, we want to expand HeartNote with multilingual support, wearable device integration, longitudinal symptom tracking, and larger medically validated datasets to improve accuracy. We also hope to collaborate with healthcare professionals and maternal health organizations to make the platform more clinically informed, accessible, and scalable for underserved communities.

Built With

css
groq
javascript
jspdf
mediapipe
nextjs
python
react
typescript
vercel
webspeech

Submitted to

VenusHacks 2026

Created by

As lead backend engineer, I designed and built the full backend architecture for HeartNote. I created both core API routes for report generation and voice transcription using Next.js App Router with Groq as our AI and transcription provider. I implemented a dual-model strategy that automatically switches to a vision-capable model when a face image is provided and falls back to text-only mode otherwise. I designed the entire type system and request/response schema to ensure type safety across the backend. I built the clinical knowledge layer including a symptom dictionary, postpartum-specific risk scoring logic, and structured prompt engineering that instructs the AI to produce SOAP-format clinical notes. I also ensured secure API key handling server-side and implemented robust error handling throughout.

Shamikta Suresh Raj
I led the UI polish and user-facing interaction layer for Heartnote, building and integrating several key components from the ground up. I developed WebcamCapture.tsx, which guides users through a preparation checklist before capturing a photo and outputs the image as base64 for the AI backend to analyze. I created the RiskBadge component with animated styling to clearly communicate LOW, MODERATE, and HIGH postpartum risk levels at a glance. To improve perceived performance during AI report generation, I built SkeletonLoader and LoadingOverlay components that keep the experience smooth while the app waits for a response. I also implemented the PDF export utility in exportPdf.ts, which captures the rendered report and generates a multi-page downloadable document users can bring to a provider. Finally, I established the app's visual design system in global.css, defining the color palette, CSS variables for risk levels, and reusable animations used across the interface.

Chloe Espenilla
As the AI / Prompt Engineering lead for HeartNote, I built the full AI pipeline behind the clinical report generation system. I wrote the two core system prompts that guide the model on postpartum cardiovascular conditions, symptom interpretation, risk classification, and structured JSON output. I also created the clinical translation layer that converts everyday symptom descriptions into medical terminology at runtime. In addition, I designed the 13-rule risk logic system that determines when symptom combinations trigger red flags and urgency warnings. On the engineering side, I wrote the shared TypeScript schema used across the entire project and built the API route connecting patient inputs, body map data, and face scan analysis to the AI model. Finally, I created scripted low, moderate, and high risk demo scenarios with a testing runner to validate outputs before the final presentation.

Likhitha Battu

Updates

Shamikta Suresh Raj started this project — May 17, 2026 12:17 PM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.