VisionBoarder - AI Vision Board Platform, Vancouver Canada
Languages
Python
Fig. 1 — AI Semantic Image Search featured view (click to expand)
01
The Problem
Business Impact
Vision board creation required manual image tagging and keyword-based search - making it impossible to find images by meaning. Users described what they wanted but couldn't find it.
Constraints
Microservice had to be stateless, horizontally scalable, and load the model once at startup for performance.
02
Technical Execution
Architectural Strategy
Built a FastAPI microservice implementing CLIP (Contrastive Language-Image Pre-training) models for high-speed semantic search across image catalogs.
Depth of Execution
User query is converted to a vector embedding. Query vector is compared against pre-computed image embeddings using cosine similarity. Supports batch processing, cross-modal search (text finds images), and multiple languages. Singleton model loading pattern for efficiency.
03
Results
✓Semantic search works across multiple languages with no manual tagging.
✓Cross-modal search - "eine Katze" finds the same images as "cat".
✓Model loads once at startup, minimizing inference latency.
Future Proofing
Containerized with Docker - horizontally scalable by spinning additional microservice instances behind a load balancer.
Project Gallery
04
Client Feedback
"I have been working with Adarsh for many years now and he always delivers a product better than I expected. He knows what the user wants and is quick to understand and implement new technologies. I have worked with dozens of coders over the past 10 years and Adarsh's experience, knowledge and service is far beyond the others."
