Orca Security and AWS Partner for Natural Language Querying

Serdar HocamAuthor & Editor

Developed to allow security analysts to easily query cloud infrastructures, the Discovery AI system has significantly reduced costs and latency.

◉ 0 views
Natural language security queries at Orca Security using Amazon Bedrock | Amazon Web Services

Orca Security, in collaboration with AWS, developed the Discovery AI agent, which enables natural language querying of cloud infrastructure data. Operating on Amazon Bedrock, this innovative system has achieved major reductions in query costs and processing times.

Natural Language Convenience in Cloud Infrastructures

It is aimed for security analysts to receive quick answers without needing deep mastery of complex query languages regarding their infrastructure. The tool developed by Orca Security covers hundreds of different cloud models.

To overcome this complexity, Discovery AI, established in partnership with AWS, translates questions asked by users as plain text into Sonar queries.

Cost and Performance Advantages

The next-generation query agent achieved a 90 percent reduction in query costs compared to the initial baseline.

Additionally, the system provided a major efficiency gain by reducing the input token volume by 93 percent and latency by 75 percent.

Nine-Stage Hybrid Architecture

Orca designed the Discovery AI system as a hybrid solution combining Retrieval-Augmented Generation with schema-aware query generation.

The system's infrastructure operates through a complex nine-stage pipeline extending from request ingestion to response post-processing.

Artificial Intelligence Models Used

The Model Selector and Generator components are powered by Anthropic Claude Haiku 4.5, while guardrails are managed with Amazon Nova 2 Lite.

Amazon Titan is used for text embeddings, whereas Cohere Rerank 3.5 re-ranks candidates to capture the highest relevance.

Continuous Development and Evaluation

The Orca team built a continuous evaluation suite and a human-in-the-loop labeling system to maintain accuracy as customer environments change.

These continuous improvement mechanisms guarantee that the security querying tool always operates up to date and with high accuracy.