Overview
What this challenge is about.
Pick a vision-language encoder (OpenCLIP, SigLIP, or BLIP-2 image-text variant). Index all 600k product images into a vector database (Qdrant/FAISS). Build a query-time pipeline that combines text embedding, optional category filter, and price filter. Construct a held-out eval set of 100 realistic queries (12 shoppers each write 8-10 queries) with relevance grading. Report recall@10, NDCG@10, and the fraction of queries where the top-1 result is judged acceptable. Compare against the current keyword baseline. Deliver a 5-page memo + the running prototype.
The Brief
What you'll do, and what you'll demonstrate.
Ship a vision-language search prototype over a 600k-SKU catalog that beats the keyword baseline on a 100-query human eval.
Earning criteria — what you'll demonstrate
- Apply vision-language models to a real retrieval problem
- Design honest human-graded retrieval evaluations
- Combine semantic and structured filters in production retrieval
- Quantify uplift over a keyword baseline that already works
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Machine Learning Engineer
Shipping vision-language retrieval at catalog scale with honest evaluation is the work MLEs do on search and discovery teams at retail and marketplace companies.
This challenge sharpens
- vision-language-models
- vector-search
- retrieval-evaluation
Computer Vision Engineer
Working with CLIP-class encoders in production and tuning retrieval over real images is exactly the work CV engineers do on AI-first product teams.
This challenge sharpens
- vision-language-models
- clip
- pytorch
AI Engineer
Building the end-to-end vector-search service plus eval harness is core AI-engineer work at companies adopting semantic search.
This challenge sharpens
- vector-search
- qdrant
- vision-language-models