HealthFlow

An end-to-end AI agent for healthcare.

Comment 1

Inspiration

We were inspired by one simple but pressing question: Why is clinical software still so painful for doctors to use? While AI has made incredible strides in consumer tech, healthcare professionals are still stuck toggling through clunky Electronic Health Records (EHRs), dictating notes manually, and wasting hours on admin that should take seconds. We saw an opportunity to build a fast, elegant AI layer that sits on top of existing workflows — something a doctor could just open in their browser and start using immediately.

Our goal was to create something that felt 10x better than the status quo: instant, voice-enabled, structured, and smart. We wanted to prove that you don’t need a 12-month procurement cycle to give clinicians something that actually works.

The Bolt Hackathon felt like the perfect place to test that hypothesis — could we go from concept to clinician-ready in under a weekend?

What it does

HealthFlow Lite is an AI-powered clinical command center for healthcare professionals. It helps doctors save time by automating their most painful admin tasks — all in a lightweight browser-based interface.

It allows clinicians to: (1) Summarise patient records into structured formats like SOAP notes or timelines using LLMs (Mistral AI) (2) Generate clinical documents (e.g. discharge summaries, clinical notes, referral letters) as downloadable PDFs (3) Transcribe voice notes (from patient consultations or dictation) using ElevenLabs, and auto-summarise them with Mistral (4) Chat with an AI assistant for clinical Q&A, including comparisons to clinical guidelines or treatment rationale (5) Auto-fill prior authorisation forms for medical insurance claims based on patient case data

How we built it

We built HealthFlow Lite as a modular React web app using Vite + TailwindCSS for fast UI development, connected to Supabase for backend services — including Postgres DB, Auth, and Storage.

We used the Mistral AI API for all language tasks, including Text summarisation, Document generation (e.g. SOAP, discharge summaries), and a Clinical Q&A chatbot.

For audio transcription, we used the ElevenLabs API, which converts uploaded or recorded voice notes into text. The output is then piped through Mistral to produce summaries or structured documentation.

We also integrated Supabase Auth for secure clinician logins with RBAC; html2pdf to export AI-generated summaries and forms; Entri for domain registration; and RevenueCat to handle subscriptions and access limits.

Challenges we ran into

One of the biggest challenges was integrating two different AI modalities — Mistral (language) and ElevenLabs (voice) — into a single, unified workflow that felt seamless for clinicians. Ensuring that voice transcriptions could be automatically summarised and linked to the correct patient case required tight control over data flow and context handling.

Another challenge was building everything to be EHR-agnostic yet clinically realistic. Since we didn’t want to rely on live hospital data or need regulatory approval, we had to simulate the core elements of a clinical workflow using a minimal Supabase schema while maintaining privacy and security principles (e.g., pseudonymisation, no write-back).

Despite these constraints, we were able to ship a fully working MVP — ensuring we make it fast, simple, and ready to deploy in real clinical sandboxes, by cutting through the fluff and noise to the core key requirements for full clinical value.

Accomplishments that we're proud of

Solo built a fully functional AI-powered clinical tool in a weekend — with working integrations across Supabase, Mistral, and ElevenLabs.

Created a realistic simulation of clinical workflows (summarising notes, transcribing voice, generating discharge summaries) without touching real patient data or relying on EHR access.

Kept the system lightweight, modular, and secure — enabling future hospitals or startups to deploy the tool in a sandbox or pilot environment with minimal setup, which is exactly what the healthcare professionals need.

Made something clinically meaningful: a tool that could actually save doctors time, reduce burnout, and improve documentation quality — with just a web browser and an API key. Something that could potentially save the US healthcare alone $1 trillion in admin spending (according to McKinsey & Co. - https://www.mckinsey.com/industries/healthcare/our-insights/administrative-simplification-how-to-save-a-quarter-trillion-dollars-in-us-healthcare).

What we learned

How to rapidly build and deploy a full-stack AI product using Supabase, Vite, and external APIs — without overengineering, and actually with minimal coding experience like myself. I have never coded an app, done Computer Science or any other technical tasks before. This was my second time building / vibe coding, and it is absolutely incredible how simple, easy and accessible Bolt has made web / app creation for everyone. I love that Bolt is democratising coding for everyone!

It was also my first time integrating an AI API (eg Mistral) into an app that I vibe coded. I have never before performed an AI integration via an API, so it was a very rewarding experience to see it work! The same goes to integrating with Supabase, RevenueCat, Entri and ElevenLabs - so it was super cool to make it work, and absolutely astonishing at how simple it actually was.

With the right tools and APIs, you can simulate meaningful healthcare infrastructure fast, and lay a foundation for real-world adoption later. Really rewarding experience!

What's next for HealthFlow

Our next step is to turn HealthFlow Lite into a real, usable product for clinicians. We’ll fully integrate RevenueCat for subscription control, run sandbox tests with clinicians using synthetic data, and refine the AI outputs for accuracy and clinical relevance. Long-term, we see HealthFlow evolving into an autonomous clinical ops assistant — automating everything from triage documentation to discharge planning, to tackle the $1 trillion spent on healthcare admin in the US alone. Maybe even apply to YC, as healthcare admin AI is on their top priority Wishlist!

Built With

  • elevenlabs
  • html2pdf.js
  • mistral
  • netlify
  • react
  • revenuecat
  • supabase
  • tailwindcss
  • typescript
  • vite
Share this project:

Updates