Why it matters
RAG cost lives in the embedding and chunking choices, not just the model call. DuckDB as an embedded vector store cuts infra overhead versus running a separate vector DB, which changes the cost math for smaller agent deployments.
The tokenmaxxing angle
Chunk size and retrieval-k directly set how many tokens get stuffed into the context window per query. A DuckDB-based pipeline is a concrete case study in trimming retrieval overhead before it ever reaches the LLM call.
From the organizers
Hosted by the Atlanta Azure Data User Group at Improving Atlanta's Alpharetta office, led by host Rob V., Monday Aug 10, 6:30-8:00pm, in-person.