BBitMonkey AI
T09The Fourteen

Prayag Infotech

Answering from documents, not from memory

Sign in to open this case study.

Sign in

The case

Prayag Infotech has 1,400 employees who ask the HR helpdesk 2,800 policy questions a month. An assistant is built to answer them from the company's own documents. The studio walks a student from a fluent invented answer to a system that gets 95% of its behaviour right - and it does that with no change to the model, no additional data and no retraining.

What is in it

  • 12 teaching steps, one animation each
  • 29 labs, checked live against the case data
  • 60 step questions, every option explained
  • 28 final paper questions, marked on submission
  • 12 Claude prompts, optional

The steps

  1. 1Yes, probably · THE PROBLEM
  2. 2Memory is not a source · THE PROBLEM
  3. 3What goes in the index · THE PROBLEM
  4. 4Chunking · FINDING IT
  5. 5Retrieval · FINDING IT
  6. 6The source rule · ANSWERING IT
  7. 7Citation, and the version trap · ANSWERING IT
  8. 8Abstention · ANSWERING IT
  9. 9The evaluation set · RUNNING IT
  10. 10The residual · RUNNING IT
  11. 11Triage · RUNNING IT
  12. 12Keeping it current · RUNNING IT
Prayag Infotech - Policy Answer Studio · BitMonkey AI