RAG ENGINE DEEP DIVE
Ground Every Answer in Relevant Context.
Control how your application finds, ranks and sends knowledge to your model.
USER QUERY
Input Intent
QUERY PROCESSING
Hypothetical Docs & HyDE
HYBRID / VECTOR RETRIEVAL
Dense HNSW + BM25
OPTIONAL RERANKING
Cross-Encoder Scorer
CONTEXT ASSEMBLY
Token Window Packing
MODEL
Configured LLM
RESPONSE + SOURCES
Verified Citations
Interactive RAG Parameter Tuning
Top K (Candidates)5 chunks
Similarity Threshold0.78
Neural Cross-Encoder Reranker
P95 retrieval latency: ~118msTest in AI Playground