Natural language security queries at Orca Security using Amazon Bedrock
Orca Security's NL-to-query AI cut token costs 93% and latency 75% on Amazon Bedrock
“Orca reduced query costs by 90 percent, input tokens by 93 percent, and latency by 75 percent compared to the initial baseline.”
Orca Security built Discovery AI on Amazon Bedrock to let security analysts query 900+ cloud infrastructure models using plain English instead of a proprietary query language. The system uses a multi-stage RAG pipeline that slashed input tokens by 93% and response latency by 75% versus a naive full-schema-in-prompt baseline. The result is a production system handling thousands of queries/day at ~15 seconds — a practical blueprint for grounding LLMs in large domain-specific schemas.