Module 08 Activity

Scenario

Users will upload files, and files are the least trustworthy input you will handle.

What you build

A file pipeline that rejects safely, chunks on sentences, extracts against a schema, and cites which part of the file an answer came from.

Steps

1. Check encoding, emptiness and binary content before any file reaches a prompt. Construct one failing example of each.
2. Chunk on sentence boundaries and measure how many chunks end at one.
3. Extract structured records with a schema that includes bounds and patterns, not just types.
4. List the qualifiers that must survive a summary and check for them mechanically.
5. Cite the file and chunk in every answer, and run a term check against the cited text.
6. Report the rejection rate by reason.

Evidence to hand in

• Three failing files with distinct reasons.
• Chunk boundary completeness at your chosen size.
• The extraction schema with bounds.
• The qualifier check on three summaries.
• An answer with citations and its term check.
• Rejection rates by reason.

Review checklist

• errors="ignore" is not used anywhere.
• Chunks end at sentence boundaries.
• The extraction schema rejects a negative amount.
• Qualifier survival is checked mechanically.
• Citations name a file and a chunk, not just a document.

Mark complete