Retrieval Quality & RAG Applications: sections and modules
Post-Retrieval Optimisation & Reranking
Design two-stage retrieval: high-recall first stage, precise second stage.
- Lessons
- 8 lessons
- By
- Sourav Banerjee
- Prerequisites
- None
Lessons
Start modulePractise and Check
- Deliverable
- Precision stack: add a cross-encoder reranker over a top-50 hybrid retrieval and measure the gain at top-5, plus the latency budget it consumes.
- Assessment
- Answer the five Interview questions without notes, then run the lab.
- Interview probe
- You add a cross-encoder reranker over the top 50 for Larkspur's support assistant. Latency per question jumps, and answers to “How long does a refund take?” are no better. Where do you look?
- Common mistakes
- Expecting a reranker to fix evidence missing from the first-stage candidates
- Loading the
CrossEncoderinsidererank_documentson every question - Keeping a chunk only when the model replies exactly
"True"
Your Progress
Saved in this browser when you leave the box.