Trademark Similarity Engine Developed a visual similarity search system for trademark infringement detection using CLIP embeddings and PostgreSQL vector search. The system encodes trademark images into 512-dimensional semantic vectors and finds visually similar marks in milliseconds. Key features: - 87-94% accuracy on trademark similarity detection - Sub-50ms query latency using pgvector IVFFlat indexing - Scales to 1M+ trademarks with approximate nearest neighbor search - Configurable similarity thresholds with automatic infringement flagging Tech: Python, CLIP (ViT-B/32), pgvector, PostgreSQL, FastAPI, PyTorch, GCP