Current research / Agents & evaluation
Agent-based playbook generation
Retrieval and testing for SOAR workflow generation
Security automation starts with an objective, but implementation requires a sequence of tool calls, conditions, and dependencies. Generating plausible code is only part of the problem.
I work on LLM-based systems for generating SOAR playbooks from high-level objectives. The work uses retrieval and contextual evidence to support generation, together with testing and iterative refinement.
This systems work connects the generation process with existing security tools and feedback from testing. It complements my research on generated test environments and on benchmarks and small-model training for playbook generation.
Research approach
- Ground generation in relevant workflow and tool context.
- Make dependencies and response requirements explicit.
- Use test feedback to inspect and refine generated behavior.