Projects | Aniket Patil

VisualSentry

Zero-shot industrial anomaly detection system that learns exclusively from defect-free images and localizes unseen defects at inference time — no defect labels required. Fuses DINOv2-powered PatchCore memory retrieval with CLIP-based WinCLIP semantic scoring, achieving 0.9545 average image AUROC across four MVTec AD categories with sub-50ms ONNX inference.

Portfolio Optimization Using Market Correlations

Forecasting stock prices using ARIMA, LSTM, and Prophet time-series models.

Production RAG System with Hybrid Search & Evaluation Framework

Production-ready RAG system using FAISS, BM25, Docker, and evaluation metrics.

Realtime Monocular Depth Mapping

Real-time monocular depth mapping using lightweight U-Net architecture.