Summary

The current codebase has no compliance checking layer. This issue tracks the implementation of the compliance checker that sits on top of the retrieval engine from #13 .

What Will Be Implemented

• Extract text from uploaded consent form (PDF or TXT)
• Auto-detect study type from consent form text using keyword matching
• Build check queue: 8 universal checks + conditional checks per detected study type
• Retrieve relevant GA4GH chunks for all checks upfront
• Single LLM call with all checks and retrieved chunks injected into prompt
• Parse structured verdicts: check name, COMPLIANT/NON-COMPLIANT, reason, citation

Builds On

#9 - document ingestion pipeline
#13 - hybrid retrieval engine