Developer struggles with PDF/document parsing for RAG in 2026

A developer running a RAG (a system that retrieves documents to help an AI answer questions) setup under contract with Canadian hospitals for non-clinical uses says their process is unreliable. Even after having review the process, the tools keep saying it's fine, but the developer knows parsing quality is poor, error-prone, and fails silently at times. They're considering subscribing to , a paid service that converts documents into AI-friendly formats, and asked the community what ingestion approaches others use.

Key points

  • Developer runs a non-clinical RAG service under contract with Canadian hospitals
  • (including Fable) reviewed the process but found nothing wrong
  • is described as inconsistent and prone to s
  • Considering a paid as an alternative
  • Asked the community how others handle PDF/PPT/docx ingestion
Read original