← learn / interactive lab
rag pipeline visualizer.
Watch a document become a grounded answer — every stage of a production Retrieval-Augmented Generation pipeline, live, inspectable, and tunable. Load the sample guide or your own PDF (≤ 5 MB) to begin.
① Ingestion — document → vector index
② Query — question → grounded answer
ask the document
chunking
600 chars
small = precise retrieval · large = more context
80 chars
prevents facts being split at boundaries
retrieval
4
chunks handed to the LLM
0.70
1.0 = pure semantic · 0.0 = pure keyword
0.25
chunks scoring below are discarded
generation
600
2000 tok
token cap for retrieved chunks in the prompt
1.0
gpt-5-mini may ignore non-default values
system prompt
applies to the next question · ingest a document first