Health-AI

Health-AI is a Flask-based personal health tracking platform that combines daily lifestyle logging, lab trend tracking, smartwatch ingestion, and evidence-based risk estimation.

What This Repo Actually Does

• User auth with local SQLite + bcrypt password hashing
• Guided onboarding for demographics, vitals, and history
• Daily logs for sleep, steps, stress, hydration, food quality, and notes
• Lab entry and historical lab timeline for 17 biomarkers
• Risk scoring with hard gating rules before any risk is shown
• OAuth smartwatch sync (Fitbit and Google Fit implemented)
• CSV import path for Apple Health style exports
• JSON API endpoints for daily history and risk summary

The app runs from app_v5.py and defaults to port 5001.

Risk Engine In This Version

The current runtime risk pipeline is rule-based and transparent, not a black-box classifier:

• CVD: ACC/AHA Pooled Cohort Equations
• Diabetes: FINDRISC score
• Cancer: lifestyle relative-risk model (not absolute incidence probability)
• Health Age: heuristic estimator based on lifestyle and labs

Implementation is in risk_calculators.py and called through compute_risk_scores() in app_v5.py.

Gating Rules Before Risk Is Displayed

compute_risk_scores() enforces:

1. Profile must include age, sex, height, weight
2. At least 7 daily logs must exist
3. At least one lab panel entry must exist
4. CVD additionally requires total cholesterol and HDL, and age 40+

If these conditions are not met, the UI returns explicit gated messages instead of fake numbers.

Data Model

SQLite DB: users.db

Tables created at startup:

• users
• daily_log
• lab_results
• surveys

Notable tracked fields include BP medication status, smoking status, alcohol/week, race, family history flags, wearable provider/token, and health goals.

Smartwatch and Import Flows

OAuth Providers Configured

OAUTH_CFG supports config keys for:

• fitbit
• googlefit
• garmin (configured in env map, but sync logic currently handles Fitbit and Google Fit)

Sync Endpoint

• GET /api/sync_watch

Behavior:

• Pulls current-day steps/active/sleep via provider APIs
• Upserts into daily_log with source='watch'

CSV Import Endpoint

• POST /import_csv

Expected columns include:

• date
• sleep
• steps
• active_minutes

Imported rows are upserted into daily_log with source='csv_import'.

ML Folder Note (ml/)

This repo also contains an advanced experimental ML stack under ml/ (predictor.py, train_model.py, safety.py, etc.).

Important: the primary runtime in app_v5.py currently uses the rule-based calculator path from risk_calculators.py, not the ml/ predictor as the default serving path.

Local Setup

python -m venv venv
# Windows
venv\Scripts\activate
# macOS/Linux
# source venv/bin/activate

pip install -r requirements.txt
python app_v5.py

Open: http://127.0.0.1:5001

Optional Environment Variables

• SECRET_KEY
• FITBIT_CLIENT_ID
• FITBIT_CLIENT_SECRET
• GOOGLE_CLIENT_ID
• GOOGLE_CLIENT_SECRET
• GARMIN_CLIENT_ID
• GARMIN_CLIENT_SECRET

Core Routes

• / home
• /register, /login, /logout
• /onboarding/<step>
• /dashboard
• /labs
• /profile
• /smartwatch
• /connect/<provider> and /oauth/callback/<provider>
• /survey
• /api/daily_history
• /api/risk_summary

Safety and Scope

This project is educational and decision-support only. It is not a medical device and is not a diagnostic substitute for clinician review.