1.5 Re-ranking Pipeline #5
Description
Goal
Build a complete re-ranking system that combines multiple signals.
Learn
- Two-stage retrieval (recall → precision)
- Feature engineering for ranking
- Linear combination vs learning-to-rank
- A/B testing ranking changes
Deliverable
- Re-ranker that combines BM25 + embeddings + business signals
- Configurable weights
- Evaluation showing improvement over single-stage
Proof Point
THIS IS THE PORTFOLIO PROJECT - you now understand every piece.
Links
→ Feeds into: FasterShops/framework Search Re-ranking Layer
Directory
search-fundamentals/05-reranking-pipeline/