This is a submission for the GitHub Finish-Up-A-Thon Challenge
What I Built
Chronic Disease Copilot is a privacy-first, voice-enabled AI health companion for people managing chronic conditions — things like diabetes, hypertension, or autoimmune disorders that require constant monitoring and coordination.
The idea came from a frustration I've felt watching people close to me struggle with the daily grind of managing a chronic illness: logging symptoms, tracking labs, preparing for doctor visits, and worrying about medication interactions. Most health apps either send your data to the cloud or feel too clinical to actually use day-to-day.
So I built something different — an AI copilot that lives entirely on your machine.
The core pillars:
- 🎙️ Voice symptom logging powered by the Web Speech API, analyzed locally by Ollama (qwen3.5 / llama3)
- 🔒 Zero Retention Mode — a HIPAA-inspired toggle where absolutely no data is written to disk
- 🧪 Local drug interaction checking and lab result explanations in plain language
- 📅 Appointment prep coach that generates doctor-visit summaries from your symptom history
The stack: React 19 + Vite + Tailwind on the front, Node.js + Express + Better-SQLite3 on the back, Ollama for all AI inference, and ElevenLabs for empathetic voice feedback.
Demo
https://youtu.be/uyzHK-c_MmA?si=r0b0TjJCtMBgc70G
The Comeback Story
This project started as a Kaggle hackathon submission. I built the core MVP in a sprint — voice input, Ollama integration, the Zero Retention toggle — submitted it, and then... left it sitting.
The README was thin. The setup was finicky. The drug interaction checker was half-baked. The appointment coach existed only as a comment in the codebase.
The GitHub Finish-Up-A-Thon gave me the nudge I needed to actually close the loop.
Here's what changed between "hackathon prototype" and "finished project":
| Area | Before | After |
|---|---|---|
| README | Basic install steps | Full setup guide, env config, usage walkthrough |
| Drug interactions | Placeholder UI | Working local AI checker |
| Appointment coach | TODO comment | Generates full pre-visit summaries |
| Zero Retention Mode | Functional but undocumented | Explained, tested, toggleable from dashboard |
| Deployment | Local only | Verified on Vercel + Google Cloud Run via Docker |
The biggest shift wasn't technical — it was deciding that "good enough for a hackathon" wasn't the same as "something I'd actually give to someone managing their health."
My Experience with GitHub Copilot
GitHub Copilot was a genuine force multiplier during the finish-up phase — less so for greenfield code and more for the unsexy completion work that makes a project real.
The moments it saved me the most:
Filling in the drug interaction logic. I had the UI shell but the actual medication parsing was TODO. Copilot suggested a structured prompt template for Ollama that extracted drug names and flagged interactions — it got the shape right on the first try and I just tuned the model instructions.
Writing the appointment summary generator. This needed to pull from symptom history, format it coherently, and produce something a doctor could actually read. Copilot accelerated the data-shaping logic significantly — the kind of "join this, group that, sort by date" code that's tedious to write from scratch.
Inline documentation. Every function I paused on, Copilot drafted a JSDoc comment that was accurate enough to ship. The README improvements were partly driven by Copilot suggesting sections I'd forgotten to cover.
Where it didn't replace my judgment: the Zero Retention Mode architecture, the privacy model, and anything touching how health data flows through the system. Those decisions needed deliberate thought, not autocomplete.
Overall: Copilot handled the mechanical completion work so I could stay focused on the parts that actually required domain thinking.
Built with Ollama · ElevenLabs · React 19 · Node.js · Vite · Tailwind CSS
Top comments (0)