_ai
I build LLM features into products and make agents useful to whole engineering teams. Staff engineer with 10+ years shipping production systems. I work spec-first, with agents, and still ship code every week.
The work below is applied AI: product features, agents, retrieval, MCP and governance. It is not model training. What ties it together is production discipline.
Prompts are built from structured data the system already has. When the data is not enough, the model says so instead of guessing.
Merges, validation and gates run on checks that behave the same way every time. The LLM advises; it doesn't decide.
LLM calls go through a provider-agnostic service with fallbacks and identity-based auth, not hard-wired API keys.
They get skills, a knowledge base and the same standards as people, so a whole team gets faster, not just one engineer.
AI 1 // _llm-feature-in-production
IoT analytics SaaS platform for HVAC and refrigeration
Technicians get grounded descriptions built from telemetry the platform already has, in the report's format.
view-case-studyAI 2 // _agent-readable-knowledge-base
An engineering organization
Agents and engineers share one source of truth that stays current, and AI review became a standing control.
view-case-studyAI 3 // _mcp-servers-and-agent-infrastructure
An engineering organization
Agents get real context without leaked keys or over-broad permissions.
view-case-studyAI 4 // _agent-skills-for-a-lending-team
Splash Financial, US consumer-lending engineering team (current role)
Shared agent skills the team uses, and delivery where every ticket traces to a committed spec.
view-case-studyAI 5 // _governance-for-ai-assisted-engineering
An engineering organization
Deterministic checks decide; the LLM advises. Merges stay gated by checks that behave the same way every time.
view-case-studyI work with Claude Code (skills, hooks, agents, playbooks, session initialization and repo-level agent instructions), Cursor and GitHub Copilot. AI-assisted review is a standing control, not an experiment.
LLM integration (OpenAI, Azure OpenAI, Anthropic SDKs), tool-using agents, MCP servers, retrieval over structured knowledge, prompt design with grounding and constrained output, Claude Code skills, hooks and agents, Cursor, Copilot, Terraform-provisioned model infrastructure, workload identity.