In one of our most exciting projects, we developed a state-of-the-art recommendation engine. Our focus was on implementing an automatic image analysis using GPT-4o. This technology enables efficient image analysis and the extraction of valuable information.
Another important aspect of the project was semantic vector search. We used OpenAI embeddings in combination with MongoDB and OpenSearch. These technologies helped to understand and process complex search queries content-wise, significantly improving the user experience.
Moreover, we tagged and indexed over 4 million customer images. This part of the project was crucial to ensure that simpleshow customers can easily and quickly find their data.
Based on this, we developed the recommendation engine. This innovative engine combines a balanced keyword search with vector search and offers advanced filtering options to make personalized recommendations even more precise and provide users with exactly the content they are interested in.
Achievements:
- Automatic image analysis with GPT-4o for over 4 million customer images
- Doubled the image recommendation quality
- Streamlining the codebase of the recommendation engine to 25% of the previous solution
- Redevelopment of the recommendation engine with a hybrid, balanced keyword and vector search + filters
- NestJS
- MongoDB
- OpenSearch
- OpenAI GPT-4o
- OpenAI Embeddings
- TypeScript
- Socket.io Websockets
- Angular
- AWS