Orca Security's New AI Lets You Ask Cloud Questions in Plain English

Orca Security, a cloud security company, has a problem: its system tracks over 900 types of cloud resources, and asking questions about them required knowing a complex query language. Analysts had to type precise model names, use 18 different operators, and navigate nested relationships. This was slow and required deep expertise. To fix this, Orca partnered with Amazon Web Services to build Discovery AI, a tool that lets you ask questions in everyday language. For example, you could type "Show me all S3 buckets that are public" and get an answer without knowing the technical details.

Discovery AI works by breaking down your question into nine steps. It first finds similar example queries to understand what you're asking, then uses several AI models from Amazon Bedrock to build a valid query. The system is smart about picking the best model for each step, which keeps costs low and speed high. Compared to their first attempt, which loaded the entire 900-model schema into one prompt, the new system is 90% cheaper, 75% faster, and uses 93% fewer tokens. It now handles thousands of queries per day with 15-second response times.

The biggest improvement came from better example retrieval. Instead of just grabbing the most similar examples, which could be near-duplicates, Orca's team tested different strategies to balance relevance and diversity. This made the AI more accurate at translating natural language into precise queries. The system also uses AWS security controls like IAM and VPC endpoints, so it meets Orca's strict security requirements.

For everyday users, this means faster answers about cloud security without needing to be a query expert. It could help companies spot misconfigurations or security risks more quickly, potentially preventing data breaches. The cost savings also mean Orca can offer this feature without passing on high expenses to customers.