Two-stage open-vocabulary vision pipeline combining OWLv2 detection with DINOv3 patch-level embeddings for zero-shot classification. Supports batched, GPU-accelerated inference, parallel object pro…

A full-stack multi-user RAG engine integrating FastAPI, PostgreSQL, Milvus, and LlamaIndex. Provides user-specific document management, incremental embedding, and metadata-based retrieval, exposed …

A clean, modular implementation of SAM2 video segmentation with point-based prompting. The pipeline preprocesses frames, extracts red/green annotation points, initializes SAM2, and propagates masks…