Natural language security queries at Orca Security using Amazon Bedrock

What happened

Orca Security created Discovery AI, a natural language query agent on Amazon Bedrock that translates plain-English questions into Sonar queries, resulting in significant improvements in cost, latency, and accuracy compared to the initial baseline.

Why it matters

For AI developers and cloud security professionals, Discovery AI simplifies querying complex cloud infrastructure data without requiring deep knowledge of domain-specific query languages, making security analysis more efficient and accessible.

• Discovery AI reduces query costs and latency by leveraging Amazon Bedrock's multi-model capabilities.
• A nine-stage pipeline with RAG and schema-aware construction improves accuracy and efficiency.
• Continuous human-in-the-loop labeling ensures ongoing system improvement and adaptability.

The bigger picture

The solution enables more efficient and cost-effective querying of cloud infrastructure data, improving security analysis workflows and reducing the barrier to entry for non-expert users.

Orca Security's Discovery AI, built on Amazon Bedrock, significantly enhances the efficiency and accuracy of querying cloud infrastructure data through a multi-stage pipeline and continuous improvement mechanisms.

Technical details

• Natural language query agent
• Amazon Bedrock
• Sonar query system
• Retrieval Augmented Generation (RAG)
• Schema-aware query construction
• Nine-stage pipeline architecture
• Anthropic Claude Haiku 4.5
• Amazon Nova 2 Lite
• Amazon Titan embeddings
• Cohere Rerank 3.5

FAQ