Overview
What this challenge is about.
You receive 80,000 anonymized product records (title, description, category, attributes) and a sample of 30,000 search log entries with click-through labels. Embed the catalog with a multilingual sentence-transformer (the retailer ships in EN/ES/PT/FR), index the lexical side in OpenSearch and the dense side in Qdrant, and design a fusion strategy (Reciprocal Rank Fusion is a fine starting point). Build an offline ranker that, given a query, returns a fused top-20. Measure nDCG@10 against the click logs vs the current BM25-only baseline. Success: nDCG@10 lift of at least 12% with a documented architecture diagram the platform team can hand to backend engineering.
The Brief
What you'll do, and what you'll demonstrate.
Design and prototype a hybrid (BM25 + dense vector) search that lifts nDCG@10 by at least 12% over BM25 on the retailer's 90-day search logs.
Earning criteria — what you'll demonstrate
- Combine lexical and dense retrieval into a single ranked list
- Apply multilingual embedding models to a real polyglot catalog
- Evaluate ranking quality with nDCG@k on click-derived labels
- Communicate an architecture trade-off (latency, cost, freshness) to non-search engineers
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.
AI Solutions Architect
Designing a hybrid retrieval architecture that survives multi-language, multi-tenant, and freshness constraints is the bread-and-butter deliverable of an AI solutions architect at any commerce company.
This challenge sharpens
- hybrid-search
- reciprocal-rank-fusion
- multilingual-retrieval
Machine Learning Engineer
Operating BM25 + dense retrieval as a measured ranking system maps directly onto how MLEs ship search and ads ranking.
This challenge sharpens
- bm25
- embedding-models
- evaluation
Data Scientist
Building click-derived labels and reporting nDCG@k by slice is the kind of measurement work data scientists own in search and recommendation teams.
This challenge sharpens
- evaluation
- hybrid-search
- multilingual-retrieval