Senior Software Engineer, Applied AI
I build LLM-powered systems from concept to production — agent workflows, tool orchestration, structured output pipelines, evaluation harnesses, and the infrastructure that keeps them reliable when real users show up. 15+ years across the full stack, now focused entirely on making AI applications production-ready and composable.
Mender — Self-healing for production agents. Built for the Google Cloud Rapid Agent Hackathon (Arize track, May–June 2026). Watches another agent's traces via the Arize Phoenix MCP server, finds regressions, generates focused evals, drafts a prompt patch, validates it against the same evals, and ships the fix to Slack with one-click approval. Also reads his own traces every cycle and tunes himself. Stack: Google ADK · Gemini 3 (Vertex AI) · Arize Phoenix + Phoenix MCP · Cloud Run · Slack. Live demo: mender-thj3gr276a-uc.a.run.app.
Beesla — An AI-native platform that converts natural language into structured workflows. Users discover relevant roles and take action through agent-driven pipelines. Built on TypeScript, Next.js, Supabase, and Docker.
LLM application architecture, agent workflows, tool calling & orchestration, RAG, evaluation frameworks (incl. LLM-as-judge), structured outputs, vector search, multi-agent systems, prompt engineering, context management, composable AI skills, agent observability + tracing, self-improvement loops.
Languages — TypeScript, JavaScript, Python, Node.js, SQL, C#
Agents & AI — Google ADK, Anthropic API, OpenAI APIs, Vertex AI / Gemini, Arize Phoenix, MCP (Model Context Protocol), vector search, RAG pipelines
Web — React, Next.js, FastAPI
Data & Infrastructure — PostgreSQL, Supabase, Redis, Docker, Kubernetes, Terraform, Firestore
Cloud — AWS, GCP (Cloud Run, Cloud Scheduler, Secret Manager, Vertex AI), Azure, Vercel
MS in Computer Engineering (USF) · Founder Institute graduate · Previously led engineering teams at PwC, Spekit, and LoanLogics, scaling orgs from 3 to 20+ engineers.