Designed a system for agents in the loop to generate versioned and parametrized artifacts for Deterministic Replications, which removes the token cost of LLM runs. Added a policy layer enforcing domain/route/action allowlists, recursive PII redaction and masking, and `human-in-the-loop` escalations for both discovery and replay runs.
GitHub: https://github.com/apervmm/llm-use-automation