RAG Office Hours — Preview and Next Cohort

RAG Office Hours: Preview and Next Cohort¶

These are example questions we tackled in office hours across previous cohorts. Office hours are not currently running.

These are representative of the practical topics we work through together.

How do we introduce time and vector search without moving to a graph database?
Is knowledge graph RAG production ready? Do you recommend it?
Would you recommend ColBERT or other specialized retrieval approaches?
Are we going to cover fundamentals of how to systematically do generation?
What’s your take on using RAG for report generation (RFPs)?
What is your experience using reasoning models as the answer generator?
How do we set user expectations on the delay with reasoning models?
How should we handle multiple RAG sources with different levels of information?
Are there open source re-rankers close to Cohere’s?
What are the best practices for chunking very long documents?
How does a long context window affect RAG systems? Does it make RAG obsolete?
What strategies can improve experimentation speed with RAG systems?
How would you evaluate parsing strategies for weird layouts, tables, and charts?
What’s the process for fine-tuning embedding models and when to use re-rankers?
What failure modes show up with synthetic data for retrieval fine-tuning?
When should we create targeted summaries vs. fine-tune embeddings?
How do you handle multilingual RAG?
How do you handle upkeep and versioning as documents change?