🧠 GroomAI — Your Personal Skincare & Grooming Advisor

A private, AI-powered skincare advisor that remembers you, analyzes your skin from photos, tracks your products, and gives honest, budget-aware recommendations tailored for Indian users.

GroomAI is not a doctor. No diagnosis, no prescriptions — just a smart, judgment-free grooming companion.

---

✨ What Makes It Different

Feature	How It Works
Long-Term Memory	FAISS vector store per user — preferences, habits, and skin details persist across sessions with auto-dedup and 90-day TTL
Vision Skin Analysis	Upload a selfie → a dedicated Vision Agent (GPT-4o-mini) returns a structured report: skin type, acne level, dark circles, texture
Product Inventory	Tracks products you already own (PostgreSQL) so the AI builds routines around them instead of recommending duplicates
Budget-Aware Search	Asks your budget first, then searches real products via Serper API with prices in ₹ and explains why each is recommended
Prompt Caching	Static system prompt + singleton agent = OpenAI's automatic prompt caching kicks in across all users, cutting ~50% token costs and reducing latency
Streaming Responses	Real-time token streaming via SSE — responses feel instant
Per-User Isolation	Every user gets isolated memory, profile, and inventory — no data leakage

---

🏗️ Architecture

┌─────────────────────────┐
│   Next.js Frontend      │  ← Auth UI, Chat, Image Upload
│   (Port 3000)           │
└────────┬────────────────┘
         │ HTTP + JWT
┌────────▼────────────────┐
│   FastAPI Backend       │  ← REST API + Streaming Chat
│   (Port 8000)           │
│                         │
│  ┌─ GroomAI Agent ────┐ │
│  │  GPT-4o-mini       │ │  ← Singleton, static prompt (cache-friendly)
│  │  8 function tools   │ │
│  └─────────────────────┘ │
│  ┌─ Vision Agent ─────┐ │
│  │  GPT-4o-mini       │ │  ← Image → structured skin report
│  └─────────────────────┘ │
└────────┬────────────────┘
         │
    ┌────▼─────┐   ┌──────────────┐
    │ Postgres │   │ FAISS + JSON │
    │ (Users,  │   │ (Memory,     │
    │  Items)  │   │  Profiles)   │
    └──────────┘   └──────────────┘

Agent Tools

The GroomAI agent has 8 function tools it can call autonomously:

• recall_memory / save_memory — semantic search over per-user FAISS vector store
• get_profile / update_profile — JSON-based user profile (name, age, gender)
• list_inventory / add_inventory_item / delete_inventory_item — PostgreSQL product inventory
• serper_search — live Google search for product recommendations

---

🔐 Authentication

Custom JWT auth (no third-party dependency):

• Register/Login → bcrypt password hashing → JWT token (HS256, 7-day expiry)
• Frontend stores token in cookie (groomai_token)
• Middleware protects routes — unauthenticated users are redirected to /sign-in
• User email is passed to the backend via x-user-email header on every request

---

⚡ Prompt Caching Strategy

OpenAI automatically caches prompt prefixes when they stay identical across requests. GroomAI exploits this:

1. Static system prompt — no dynamic values (no user ID, no date) baked into the instructions
2. Singleton agent — built once at import time, reused for every user and every request
3. Session context injection — dynamic values (date, user ID) are injected as the first user message in the conversation, after the cached system prompt

Result: ~50% reduction in input token costs and lower latency for all users.

---

🐳 Local Development (Docker)

The full stack runs with a single command:

# 1. Clone and configure
git clone https://github.com/IshantSingh24/GroomAI.git
cd GroomAI
cp .env.example .env
# Edit .env → add your OPENAI_API_KEY and SERPER_API_KEY

# 2. Start everything
docker compose up --build

This starts:

Service	URL	Live Reload
Next.js frontend	http://localhost:3000	✅ Source mounted
FastAPI backend	http://localhost:8000	✅ --reload + source mounted
PostgreSQL 16	Internal (port 5432)	—

Code changes in backend/ or frontend/ reflect instantly without rebuilding.

Environment Variables

Variable	Required	Description
OPENAI_API_KEY	✅	Powers the AI agents and embeddings
JWT_SECRET	✅	Signs JWT tokens — use a long random string
SERPER_API_KEY	For search	Enables live product search via Google Serper
DATABASE_URL	Auto-set	Docker Compose sets this automatically

---

🚀 Deployment

• Backend → Render (configured via render.yaml)
• Frontend → Vercel (standard Next.js deployment)
• Database → Render managed PostgreSQL (injected via DATABASE_URL)

---

🛠️ Tech Stack

Layer	Technology
Frontend	Next.js 16, React 19, TailwindCSS
Backend	FastAPI, Uvicorn, Python 3.12
AI	OpenAI Agents SDK, GPT-4o-mini, text-embedding-3-small
Memory	FAISS (vector), JSON (profiles)
Database	PostgreSQL 16 (users, inventory)
Auth	JWT (python-jose) + bcrypt
Search	Google Serper API
Infra	Docker Compose (local), Render + Vercel (prod)

---

⚠️ Disclaimer

GroomAI is an informed advisor, not a medical professional. It does not diagnose conditions, prescribe treatments, or replace dermatologist consultations. If a concern seems serious, it will advise you to see a specialist.