AI in production
That's what Root Cause, the AI feature my team shipped for Syngenta, analyzes in production every week. I've also built AI report pipelines, rolled out coding agents across an engineering team, and run local models on my own hardware to understand the tradeoffs firsthand.
The work
Led delivery of Root Cause for GHX Fields at Syngenta North America. It analyzes agronomic and environmental patterns across 90,000 fields a week and surfaces concerns with suggested mitigations — before the grower has to ask.
Built report-generation pipelines where LLM output is grounded in operational data and reviewed before it reaches a client. Narrative quality matters less than being checkably right.
Rolled out AI-assisted development at Skyward — coding agents for senior engineers on large codebases, with the same code-review bar as human-written changes.
Run hosted-versus-local comparisons with Ollama and vLLM. For some client data, the privacy answer decides the architecture before cost or latency get a vote.
How Root Cause works
The AI sits in the middle of an existing agronomic workflow. It gets growers to a problem earlier; the decision about what to do stays with people who know the field.
Principles
Open source
Small, working tools from my own AI workflows — agent task boards, image-generation wrappers, local-model experiments.
Kanban-style control surface for managing AI agent work alongside an existing toolchain.
Node CLI wrapping Codex image generation for repeatable asset workflows.
Notes and experiments from running models locally with vLLM and Ollama.
REST middleware that lets API clients shape response payloads, cutting endpoint sprawl.
