~/divyam/stories groomme-ai-recommendation-engine
ai junction · ai systems
GroomMe: An AI Recommendation Engine
A full-stack recommendation platform that uses Django REST Framework and Gemini to turn wardrobe data into context-aware outfit suggestions.
- when
- August 2025
- role
- Full-Stack Engineer
- stack
- DjangoDRFGemini APIPostgreSQLJWTPython

Overview
GroomMe is a backend-first recommendation engine for personalized outfit suggestions. Instead of relying on simple filters, the system uses structured wardrobe data and user preferences to give the model grounded context.
The platform is built with Django REST Framework, JWT-secured endpoints, and a Gemini API recommendation layer designed around prompt conditioning.
Problem
Generic outfit recommendation systems often break down when a user's wardrobe is small or their taste is nuanced. The challenge was to make recommendations feel personal without needing a massive collaborative-filtering dataset.
Solution
I modeled wardrobe items and style preferences as normalized metadata, then passed only the relevant attributes into Gemini. That keeps the recommendation grounded in what the user owns instead of letting the model drift into generic styling advice.
Authentication is handled with SimpleJWT, while custom request-validation middleware helps protect user data across the API surface.
Highlights
- DRF backend with normalized wardrobe and preference schemas
- JWT-secured APIs with request validation and logging middleware
- Gemini API integration with prompt conditioning and attribute filtering
- Context-aware outfit suggestions grounded in user-owned wardrobe items
Outcome
The project showed that LLM-driven personalization can work well in a constrained domain when the input context is carefully structured. The prompt design and metadata model carried more value than a large training pipeline would have at this stage.