About
From client operations to LLM systems that hold up in production
I spent 10 years in client-facing operations before moving into AI engineering. That background is why I care about the unglamorous parts of LLM systems: what happens when the model is wrong, when a lead writes in Spanish, or when a prompt change quietly drops accuracy.
My paid work so far: an SMS lead-qualification deployment for Acuity Real Estate, document extraction for DocExtract, and private AI application work for EnterpriseHub. Alongside that I run agent-security research on how tool output can hijack an agent, and I publish the misses as well as the hits.
What I want next: a full-time AI Engineer role on a team that ships LLM features to real users and wants them measured and gated, working US remote.
What I'm good at
- Python and FastAPI services around LLMs: extraction, retrieval, tool use, structured output
- Evals as CI checks: fixture replays, adversarial sets and score floors, with LLM-as-judge in paid live jobs
- Agent and MCP security: indirect prompt injection, tool-dispatch design, auth on tool servers
- Turning vague requirements into scoped, testable slices with a clear stop condition
Roles I'm targeting
- AI Engineer / Applied AI Engineer
- AI Backend Engineer / LLM Platform Engineer
- Forward Deployed Engineer, where the job is still building and shipping
- Full-time, US remote
Stack
LLMs and orchestration
Claude API, OpenAI API, Gemini API, MCP, RAG, agent workflows, tool use, structured output, prompt caching
Evaluation and safety
pytest, offline fixture replays, RAGAS, LLM-as-judge, Promptfoo, adversarial and prompt-injection sets, pre-registered experiments
Backend and data
Python, FastAPI, PostgreSQL, pgvector, Redis, ARQ, Docker, BM25 and hybrid retrieval, GoHighLevel CRM integration
Delivery and operations
GitHub Actions CI, OpenTelemetry, structured logging, Terraform and Kubernetes manifests, runbooks and handoff docs