🩺 Metabolic Care Assistant

AI-powered dietary support and blood sugar tracking for people living with diabetes — built with real-world clinical safety in mind.

---

🌍 The Problem

Over 33 million Pakistanis live with diabetes — the 3rd highest burden in the world. Millions more are undiagnosed. Yet access to personalised dietary guidance from clinicians is limited, expensive, and time-consuming.

Most people managing diabetes are left asking:

• "Can I eat biryani?"
• "How much roti is too much?"
• "Is my blood sugar stable enough to change my habits?"

Generic apps don't answer these questions in the context of desi food culture. Clinicians don't have time to give meal-by-meal guidance. There's a gap — and this tool fills it.

---

💡 What It Does

The Metabolic Care Assistant is a Generative AI-powered web app that:

1. Triages users safely — a clinical safety layer (GREEN / AMBER / RED) determines whether the user's blood sugar and BP readings are stable enough to use the tool independently, or whether they need to see a clinician first.
2. Generates a personalised 7-day meal plan — tailored to desi food preferences, vegetarian options, hypertension, and high cholesterol.
3. Logs and visualises blood glucose readings — with trend charts and stability insights.
4. Tracks daily plan adherence — and uses AI to coach users when they deviate from the plan.
5. Suggests healthy food swaps — using an LLM to provide safe, non-medical, practical alternatives.

---

🧠 Generative AI Integration

The app uses Groq's API with LLaMA 3.3 70B for two core AI features:

1. Healthy Swap Suggestions

When a user views their meal plan, they can request AI-generated alternatives. The LLM receives the day's meals and returns 3 culturally relevant, diabetes-safe swap ideas — no medication advice, no diagnosis, just practical food guidance.

User: "Day 3 plan includes nihari + naan"
AI: → "Try a smaller naan portion with added salad"
    → "Replace oil-heavy nihari gravy with a lighter shorba version"
    → "Add a side of raita (unsweetened) for protein without extra carbs"

2. Deviation Coaching

When a user logs that they didn't follow the plan and describes what they ate, the LLM gives 3 personalised tips on how to handle that food choice better next time — portion control, carb awareness, preparation methods.

Both AI calls are:

• Strictly non-diagnostic (enforced via prompt engineering)
• Gracefully fall back to sensible defaults if the API is unavailable
• Structured as JSON arrays for reliable parsing

---

🏗️ Architecture

┌─────────────────────────────────────────────────────┐
│                   Streamlit Frontend                │
│  Login → 7-Day Plan → Profile → Check-in → Dashboard│
└──────────────────┬──────────────────────────────────┘
                   │
        ┌──────────┴──────────┐
        │                     │
┌───────▼──────┐    ┌─────────▼────────┐
│  Groq / LLaMA│    │ Supabase Postgres│
│  3.3 70B     │    │  (via SQLAlchemy)│
│  - Swaps     │    │  - profiles      │
│  - Coaching  │    │  - glucose_logs  │
└──────────────┘    │  - daily_checkins│
                    └──────────────────┘
        │
┌───────▼──────────────────────────────┐
│           Core Python Modules        │
│  triage.py   — safety routing logic  │
│  planner.py  — meal plan generation  │
│  meal_bank.py— curated meal dataset  │
│  storage.py  — DB abstraction layer  │
│  llm.py      — Groq API integration  │
│  config.py   — thresholds & settings │
└──────────────────────────────────────┘

Privacy by design

Phone numbers are never stored. A one-way SHA-256 hash (with a secret salt) is used as the user identifier. Even if the database was compromised, no phone numbers could be recovered.

---

🔐 Clinical Safety Layer (Triage)

Before the user accesses any features, the app evaluates:

Signal	GREEN	AMBER	RED
HbA1c	< 8%	8–9%	≥ 9%
Fasting glucose variability	Low std dev	Std dev ≥ 25	Std dev ≥ 45 or range ≥ 120
Systolic BP	< 140	140–179	≥ 180
Diastolic BP	< 90	90–119	≥ 120
Total cholesterol	< 200	200–239	—
Other major conditions	—	—	Immediate RED

RED users are blocked from all features and directed to seek clinician care. This is a deliberate, non-negotiable safety decision.

---

🗂️ Project Structure

metabolic-care-assistant/
├── app.py           # Main Streamlit app — UI and flow
├── storage.py       # Database layer (SQLAlchemy + Supabase/SQLite)
├── triage.py        # Clinical safety routing logic
├── planner.py       # 7-day meal plan generation
├── meal_bank.py     # Curated meal dataset (desi + international)
├── llm.py           # Groq/LLaMA AI integration
├── config.py        # Thresholds, carb constants, app settings
├── requirements.txt # Python dependencies
└── .streamlit/
    └── secrets.toml # API keys (not committed — see setup below)

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• A Groq API key (free tier works)
• A Supabase project (free tier works) — or runs on local SQLite automatically

1. Clone the repo

git clone https://github.com/umer1556/metabolic-care-assistant.git
cd metabolic-care-assistant

2. Install dependencies

pip install -r requirements.txt

3. Set up secrets

Create .streamlit/secrets.toml:

GROQ_API_KEY   = "your-groq-api-key"
DATABASE_URL   = "postgresql+psycopg://USER:PASSWORD@HOST:PORT/DBNAME"
PHONE_SALT     = "any-random-secret-string"

# Optional — defaults shown
GROQ_BASE_URL  = "https://api.groq.com/openai/v1"
GROQ_MODEL     = "llama-3.3-70b-versatile"

Where to find DATABASE_URL: Supabase dashboard → Project Settings → Database → Connection string → Transaction pooler tab. Copy that string exactly.

No Supabase? Leave DATABASE_URL empty — the app automatically falls back to a local data.db SQLite file with no setup required.

---

4. Set up the database (Supabase only)

If using Supabase, run the following in the SQL Editor (Supabase dashboard → SQL Editor → New query):

-- Drop old tables if they exist (clean slate)
DROP TABLE IF EXISTS public.daily_checkins;
DROP TABLE IF EXISTS public.glucose_logs;
DROP TABLE IF EXISTS public.profiles;

-- Profiles table
CREATE TABLE IF NOT EXISTS public.profiles (
    user_key        VARCHAR(80) PRIMARY KEY,
    full_name       TEXT,
    phone_last4     VARCHAR(8),
    age             INTEGER,
    gender          TEXT,
    height_cm       INTEGER,
    weight_kg       NUMERIC,
    family_history_json TEXT,
    diabetes_type   TEXT,
    has_hypertension    INTEGER,
    has_high_cholesterol INTEGER,
    created_at      TIMESTAMPTZ DEFAULT NOW(),
    updated_at      TIMESTAMPTZ DEFAULT NOW()
);

-- Glucose logs table
CREATE TABLE IF NOT EXISTS public.glucose_logs (
    id              BIGSERIAL PRIMARY KEY,
    user_key        VARCHAR(80) NOT NULL,
    measured_at     TIMESTAMPTZ NOT NULL,
    logged_at       TIMESTAMPTZ DEFAULT NOW(),
    reading_type    TEXT NOT NULL,
    value           NUMERIC NOT NULL,
    meal_note       TEXT
);

-- Daily check-ins table
CREATE TABLE IF NOT EXISTS public.daily_checkins (
    id              BIGSERIAL PRIMARY KEY,
    user_key        VARCHAR(80) NOT NULL,
    checkin_date    DATE NOT NULL,
    followed_plan   INTEGER NOT NULL,
    actual_meals    TEXT,
    created_at      TIMESTAMPTZ DEFAULT NOW()
);

Note: The app will also auto-create these tables on first run via init_db() — the SQL above is provided for transparency and manual setup.

5. Run the app

streamlit run app.py

---

☁️ Deploying to Streamlit Cloud

1. Push your code to GitHub (ensure .streamlit/secrets.toml is in .gitignore)
2. Go to share.streamlit.io → New app → select your repo
3. Add your secrets in the Streamlit Cloud dashboard under Settings → Secrets
4. Deploy

---

📦 Dependencies

Package	Purpose
streamlit	Web UI framework
sqlalchemy>=2.0	Database ORM
psycopg[binary]	PostgreSQL driver (psycopg3)
openai>=1.0	Groq API client (OpenAI-compatible)
pandas	Data handling for dashboard
matplotlib	Glucose trend charts

---

🍽️ Meal Bank

The current meal bank contains 14 curated meals across breakfast, lunch, dinner, and snacks — all tagged for dietary properties:

Tag	Meaning
desi	South Asian dishes
veg	Vegetarian
high_fiber	Good for blood sugar stability
low_sodium	Suitable for hypertension
low_satfat	Suitable for high cholesterol

The planner filters meals by these tags based on the user's conditions — e.g. a hypertensive user will only receive low_sodium meals where available.

The meal bank is designed to be easily expanded. Each meal is a Python dict — a nutritionist or clinician can add entries without touching any app logic.

---

⚠️ Disclaimers & Limitations

• Not a medical device. This is an educational prototype built for a hackathon.
• Not medication advice. The app never suggests insulin doses, medication changes, or diagnostic conclusions.
• Not a replacement for clinical care. Users with RED triage results are explicitly blocked and directed to a clinician.
• The meal bank is small (14 meals) and should be reviewed by a registered dietitian before any real-world deployment.
• AI-generated suggestions are validated against a strict prompt but are not reviewed by a clinician in real time.

---

🔮 Future Roadmap

• Food photo analysis (multimodal) — snap a meal, get instant carb estimate
• Integration with glucose meters via Bluetooth / manual CSV import
• RAG over clinical dietary guidelines (e.g. ADA, IDF) for evidence-backed suggestions
• WhatsApp bot interface for low-tech users
• Dietitian review portal — flag high-risk users to a clinician dashboard
• Expanded meal bank with 100+ desi dishes, validated by a nutritionist

---

👥 Team

Built at the HEC Generative AI Hackathon by:

Name	GitHub
Muhammed Umer	@umer1556
Zunaira Hawar	@zunairakhan123
Jaisha Khan	@jaishakhan
Muhammad Husnain	@MuhammadHusnain572
Dr.Fatima Farhat	@docffarhat-prog

---

📄 License

MIT License — see LICENSE for details.