Building an AI Skincare Platform with Computer Vision & E-Commerce
Full Stack Engineer (Solo)
SkinGenii is an AI-driven dermatological e-commerce platform that personalizes skincare recommendations through computer-vision analysis, ingredient intelligence, and quiz-based profiling. I built the entire platform solo — from database schema design to AI integration to payment processing.
The Problem
The skincare industry suffers from information asymmetry — consumers can't decode INCI ingredient lists, recommendations are one-size-fits-all based on popularity not skin parameters, and professional assessments cost ₹500–2000 per session. Users are left guessing about their skin type and concerns.
Consumers can't understand chemical ingredient nomenclature
Product recommendations ignore individual skin parameters
Professional skin assessments are expensive and appointment-gated
No way to verify if a product's ingredients are safe for sensitive skin
The Approach
Deploy AI-at-the-edge: a quiz + photo analysis pipeline that produces dermatologist-grade scores and uses them as a filtering layer over a product catalog with multi-dimensional scheduling metadata.
AI Skin Analysis
Google Gemini 1.5 Flash analyzes facial photos to generate scores across 14 dermatological dimensions (blemishes, wrinkles, pores, oiliness, acne, etc.).
Scheduling Matrix Engine
Products have customSchedule arrays with suitability indices per skin parameter — enabling O(1) matching via MongoDB $elemMatch queries.
Ingredient Safety Intelligence
OCR-powered ingredient scanning with a database of 91 classified cosmetic functions and EWG/CMG hazard ratings.
Full E-Commerce Stack
Razorpay integration with HMAC-SHA256 verification, order lifecycle state machine, automated email notifications via MSG91.
System Architecture
Interactive architecture map — click any node to trace its connections.
Data Flows (click a node to filter)
The Build
Solo-built from zero to production, handling every layer from AI integration to payment security.
Foundation
Next.js 14 App Router setup, MongoDB schemas for 11 collections, NextAuth JWT authentication with role-based middleware.
AI Pipeline
Integrated Gemini 1.5 Flash with structured JSON output schema, built fallback to OpenAI GPT-4o-mini, added Tesseract.js OCR for ingredient label scanning.
Recommendation Engine
Designed the scheduling matrix architecture — each product maps suitability across 14 skin parameters. AI scores transform directly into MongoDB filter predicates.
E-Commerce & Payments
Razorpay order creation + cryptographic verification, order status state machine (Pending→Processing→Shipped→Delivered), automated transactional emails.
What Was Built
A complete AI skincare platform with quiz-based profiling, photo analysis, ingredient safety checking, product recommendations, shopping cart, payments, and a full admin panel.
Dual-AI skin analysis: Gemini 1.5 Flash (primary) + OpenAI GPT-4o-mini (fallback)
Product scheduling matrix enabling instant personalized recommendations
OCR + NLP ingredient extraction with 91-function safety classification
Complete e-commerce: cart, checkout, Razorpay, order lifecycle, email notifications
The Results
A live production platform at skingenii.com serving real users with AI-powered skincare intelligence.
Lessons Learned
“Structured AI output > free-form parsing”
Forcing Gemini to return responseMimeType: 'application/json' with a strict schema eliminated 100% of parsing failures vs free-text responses.
“Scheduling matrices beat collaborative filtering”
For skincare, explicit dermatological parameter matching is more accurate than user-behavior-based recommendations. The matrix approach is also debuggable.
“Multi-layer validation prevents bad data”
Zod schemas → Mongoose type constraints → pre-save hooks → business logic guards. Four layers mean bugs are caught early and consistently.
Tech Stack
Frontend
Backend
AI/ML
Database
Payments
DevOps
Digitizing a Non-Profit's Financial Operations from Paper to Production
BSRST • Sep 2023 – Present