TL;DR
- Problem
- E-commerce is flooded with fake reviews.
- Built
- Implemented a hybrid architecture.
- Result
- 94% accuracy · 10+ platform coverage
01The problem
Developed an AI-driven browser extension that detects and filters fake reviews in real time, integrating seamlessly with major e-commerce platforms to ensure trustworthy online shopping experiences.
E-commerce is flooded with fake reviews. The challenge was creating a real-time detection system across multiple platforms without degrading browser performance.
02What I built
Implemented a hybrid architecture. A local TensorFlow.js model runs in-browser for instant scoring, while complex cases are offloaded to a DistilBERT backend, ensuring 99% uptime.
- Dual-Engine AI Detection: Hybrid local (TensorFlow.js/ONNX) and backend analysis for maximum accuracy and speed
- Lightning-Fast Performance: Optimized with Vite bundling, lazy loading, and efficient DOM manipulation for zero lag
- Customizable Experience: Dark/light themes, adjustable highlight colors, auto-analyze mode, and granular privacy settings
- Enterprise-Grade Security: Client-side processing option for privacy-conscious users with optional backend enhancement
03The hard part
Scores in the browser first (TensorFlow.js); only hard cases hit the DistilBERT backend. Speed and privacy over peak accuracy.
04Results
- Accuracy
- 94%
- Platform Coverage
- 10+
- Reviews Analyzed
- 10k+
05Stack
- React
- TypeScript
- TensorFlow.js
- ONNX Runtime
- Vite
- DistilBert
- FastAPI