Personal Project · 2025–26

AI health coach

Designed and shipped a production-grade AI health coaching app — pulling real biometric data, generating personalised coaching via LLM, and deploying it as a mobile PWA.

Google Health API Llama 3.3 via Groq Google Apps Script Cloudflare Pages PWA No external developers
Tue, 9 Jun
Today
Coach
Profile
73 WELLNESS
1,183
Steps
28m
Active
7.3h
Sleep
44ms
HRV
86%
SpO2
60bpm
HR
😴 SleepGood
7.3 hrs
⚡ Active zone56%
28 min

6
biometric signals tracked in real time
70B
parameter LLM powering the coach
£0
monthly infrastructure cost

The Product

A health coach that knows
your actual data.

Most health apps show you charts. SGHG tells you what to do about them — with specific, data-driven recommendations generated fresh each morning.

📡
Live biometric ingestion
Pulls steps, sleep stages, HRV, heart rate, SpO2, and Active Zone Minutes daily via the Google Health API. No manual entry, no third-party sync.
Google Health API
🧠
AI morning briefing
Every morning, a Llama 3.3 70B model reads the last 7 days of health data and your personal profile to generate a specific briefing: trends, today's focus, and Indian vegetarian diet recommendations.
Llama 3.3 via Groq
💬
Conversational health coach
A full chat interface with conversation history, quick-reply chips, and typing indicators. The coach has context of your health data and past conversations — not a generic chatbot.
Context-aware LLM
Activity rings + wellness score
An Apple Watch-style ring dashboard with an AI-computed wellness score at the centre. Each ring represents a metric relative to your personal goal, not population averages.
Custom scoring model
📱
Installable mobile PWA
Deployed on Cloudflare Pages and installable to the iPhone home screen. Full-screen, no browser chrome, no App Store dependency. Works offline for cached data.
PWA · Cloudflare Pages
🔒
Private by design
All health data stays in a private Google Sheet. No third-party health data aggregators, no advertising SDKs. The AI API calls include only the minimum context needed.
Zero data sold

The Architecture

Designed for zero running cost.

The entire backend runs on Google Apps Script — a serverless environment that triggers on a schedule, pulls from the Google Health API, and writes to a Google Sheet acting as the database.

The AI layer calls Groq's inference API with Llama 3.3 70B — fast enough for real-time chat responses with a sub-2 second latency on most queries.

The frontend is a single HTML file deployed to Cloudflare Pages, calling the Apps Script endpoint via GET requests. No Node.js, no build step, no CI/CD pipeline to maintain.

Google Apps Script Google Health API v4 Google Sheets (DB) Groq API Llama 3.3 70B Cloudflare Pages OAuth 2.0 PWA
Fitbit / Tracker wearable device Google Health API steps · sleep · HRV · SpO2 · HR · AZM Google Apps Script scheduled · serverless · doGet/doPost Google Sheets DailyLog · Briefing · Profile Groq · Llama 3.3 70B · <2s latency Cloudflare Pages static PWA · no server · free tier iPhone PWA