Nexus
Search an image library by describing what is in the picture, using CLIP embeddings and Elasticsearch.
- 2025
- Team project
- Semantic image search
Overview
Nexus finds images from a description instead of relying on filenames or tags. CLIP turns both the images and the search text into vectors in the same space, so a text query can be matched against pictures directly.
Elasticsearch handles retrieval, MongoDB stores the metadata and Cloudinary serves the media. A FastAPI service runs the model and a Next.js front end sits on top.
Everything runs under Docker, so the model service, the search index and the database come up together.
What I built
- Built the FastAPI service that generates and serves CLIP embeddings
- Set up the Elasticsearch index and the vector retrieval
- Built the Next.js search interface
- Wired Cloudinary for media and MongoDB for metadata
- Containerised every service with Docker
Medical Chatbot