Projects

AI systems, RAG pipelines, and full-stack products I've built along the way.

Transformer-Based WAF Pipeline cover

Transformer-Based WAF Pipeline

A reverse-proxy WAF combining sliding-window rate limiting with transformer-based request inspection, benchmarked against ModSecurity with OWASP CRS v4 on 675 labeled real-world payloads for a 96% detection rate vs. 46%. Fine-tuned DistilBERT to detect 10 web attack classes - Path Traversal, SQLi, XSS, SSRF, and IDOR - at 97.5% accuracy, with OWASP Juice Shop, WebGoat, and DVWA deployed behind a Docker Compose reverse-proxy gateway enforcing per-IP rate limiting and monitor, block, and challenge modes before reaching the origin.

PyTorch
DistilBERT
ONNX
FastAPI
Next.js
Docker
Redis
Nginx
Deepfake Detection System cover

Deepfake Detection System

Full-stack AI platform with a 4-agent orchestration system (Perception, Detection, Compression, Cognitive) for deepfake media analysis. ResNet50/EfficientNet models trained on FaceForensics++ (16GB, 7000+ samples) for high-accuracy detection of AI-generated content.

Next.js
React
TypeScript
Node.js
MongoDB
FFmpeg