Open-source project
TwinzyAI
A privacy-focused AI style and vibe game that reads visible traits from a photo without ever identifying anyone.
Overview
Players upload a photo and get a playful reading of their visible style and vibe, which they can share through a temporary link. The whole product is designed around a constraint: it must be fun without becoming a face-recognition tool.
Architecture
A Next.js front end with a NestJS and Fastify API, containerised with Docker, calling Gemini for analysis. The pipeline validates the image, extracts visible traits, discards the image buffer immediately, and every stage after that works on text only.
Engineering
The privacy properties are the feature: no facial recognition, no identity matching, no image persistence, and text-only processing once traits are extracted. Around that sit secure image validation, safety gates on generated content, expiring share links, bilingual Arabic and English support, and a mobile-first PWA build.
- Role
- Built the full stack and designed the privacy pipeline that keeps image data out of storage entirely.
- Category
- AI
- Stack
- Next.jsNestJSFastifyTypeScriptGeminiDocker
- Language
- TypeScript