At PromptLedger, I built an AI behavior control plane for versioning, evaluating, and monitoring AI behavior. It includes an A/B evaluation harness, a gate that blocks unevaluated prompts, and run tracing with drift detection and token metering.
I also built an AI Operations Copilot spanning five departments and five production AI tools. Its security-first RAG filters role permissions inside the vector query, with validation through unit tests, an RLS matrix, and agent-behavior evaluations.
In selected engineering projects, I developed an enterprise RAG and agent platform with hybrid retrieval, bounded LangGraph agents, human-in-the-loop approvals, and an evaluation harness. I also created an MCP server that traces dependency paths to vulnerabilities using a Neo4j knowledge graph.
My other projects include a metered multi-provider LLM gateway with fallback chains and circuit breaking, local LLM serving on NVIDIA GPUs, and a voice event layer that extracts speech into schema-constrained events. At R2s, I analyzed BPMN 2.0 processes and user stories, translating operational requirements into implementation-ready workflows.
