AI / ML / 2025
PixelSearch
Find the picture in your mind. Natural-language search for visual collections.
The problem
Image libraries grow faster than their labels. Filename search misses the subjects, scenes, and associations people actually remember.
My role
Embedding pipeline, retrieval logic, model integration.
Semantic image retrieval combining vision-language embeddings, indexing, and captioning.
The approach
- Represented images and text with CLIP embeddings in a shared vector space.
- Used FAISS for similarity search and explored BLIP-2 captioning to enrich image context.
- Investigated FaceNet-based facial recognition as a separate retrieval experiment.
The toolkit
The outcome
Built a retrieval workflow that connects descriptive language to images without relying only on filenames.
No public repository, demo, or benchmark is supplied in the portfolio source. The preview illustrates the concept, not a shipped interface.
Keep exploring