From language to a bounded query interface
Thesis: Orca’s case illustrates how a model interface becomes operationally useful when it emits a structured domain query rather than a free-form answer. The system translates natural-language questions into Sonar filters, making the model output an input to an existing security data interface.
Observed facts: Sonar covers more than 900 cloud infrastructure and application resource models and supports 18 operators. The initial proof of concept loaded the full schema into one prompt, using roughly 2 million input tokens per workload and taking about 60 seconds; AWS says the production pipeline handles thousands of queries daily at about 15 seconds. Mechanism: Orca combines retrieval-augmented generation with schema-aware query construction in a multi-stage pipeline, matching stages to different Bedrock models. This narrows the output contract from an open-ended response to a query that can be applied to Sonar.
Why now: The case addresses a practical interface bottleneck: analysts need to explore a complex cloud graph without learning its schema and query operators first. Implication: The useful design move is to treat the generated query as a task interface whose validity can be evaluated against the receiving system, rather than treating fluency as success. AWS reports 90% lower query cost, 93% fewer input tokens, and 75% lower latency versus Orca’s initial baseline. Evidence boundary: These are figures in an AWS customer case study, comparing the deployed approach with the initial proof of concept; they are not independently audited results, and the supplied account does not establish production accuracy rates. Watch condition: Look for published query-validity and correction rates, plus evaluation across complex multi-model questions, to determine whether the efficiency gains preserve reliable results.