Bug: LLM Extracts Incorrect or Null Values from PDFs #43
Description
The LLM is struggling to correctly extract key fields from useful PDFs, returning null or inaccurate values for fields such as species name, stomach counts, study location, and study date. This directly affects the quality and reliability of the pipeline output.
Possible Causes:
- Too much irrelevant text being passed to the LLM
- Prompt wording may not be specific enough for certain paper formats
- LLM may be struggling with tables or non-standard formatting
Proposed Next Steps:
- Broaden and refine prompt wording with additional field hints
- Implement auto-retry logic when key fields return null
- Test against a labeled set of PDFs to measure improvement