Document generation hits quality problems at 10,000-file scale
The plan is to place all available data in a RAG-based knowledge store and use LLMs to create presentations, PDFs, and assignments. The generated files currently have poor quality, including layout and display errors. Images also need to be included, but no way to improve the results has been identified.
The intended scale is more than 10,000 documents. No model, document-generation tool, , cost, , or test result is specified.
Key points
- All available data would be collected in a RAG-based knowledge store.
- LLMs would generate presentations, PDFs, and assignments.
- The current files have layout and display problems.
- Images must also be included in the generated documents.
- The target exceeds 10,000 documents, but cost and are unknown.