Weekend Challenge · Week 3
The Chunking Bake-Off
Everyone copies chunk_size=1000, overlap=200 from a tutorial. You're going to find out what those numbers should be for YOUR corpus.
Your mission
- Run at least 4 chunking configurations across size, overlap, and structure-awareness
- Score each configuration on your full query set: Recall@5 and MRR
- Inspect 3 losing retrievals per configuration: WHY did that chunking lose?
Deliverable
A bake-off table with metrics per configuration and a written recommendation for your capstone, committed.
Stretch goal
Design one corpus-specific chunking rule (split on headings, keep tables whole, attach section titles to every chunk). Does it beat all generic configurations?