Software Engineer · AI Agent Tooling & Security
I build tooling for AI agents — the layer that makes autonomous systems observable, testable, and reliable — plus an always-on Windows security agent. Two focused areas, no side quests.
A focused toolchain for building and operating AI agents: write patterns → diagnose the runs → observe behavior.
| Project | What it does | Stack |
|---|---|---|
| agent-eval | Zero-dependency toolkit for evaluating AI agent outputs. Post-hoc and report-only: it reads the traces an agent already produced, finds the runs that failed the process — timed out, ran away, stalled, or lied about their own outcome — worst-first, and hands back the diagnostic data. It reduces the search space for a fix; it never gates a build. (diagnose the runs) | TypeScript |
| AgentLens | Observability & explainability for AI agents — trace how an agent reached its answer: capture every model + tool step, debug, and understand behavior. SDK + collector + dashboard. (observe the runs) | Python |
| prompt | Prompt engineering toolkit — template engine, prompt chaining, injection detection, and bias checks. (write the patterns) | C# / .NET |
| agentic-recipes | Canonical agentic pipeline patterns — summarization chains, tool-agent ReAct loops, RAG, plan-and-execute, self-consistency, multi-agent debate, reflexion, tree-of-thoughts. (compose the patterns) | C# / .NET |
| Project | What it does | Live |
|---|---|---|
| WinSentinel | Always-on Windows security agent — real-time monitoring, AI-powered correlation, auto-remediation, and audit modules with compliance profiles. | winsentinel.ai |
| WinSentinel Pro | Pro tier — license server, fleet management, and advanced modules for organizations. | Site |
Building tools that make intelligent systems observable, accountable, and reliable.