What I build

LLM features that do a specific job inside a product that already exists: document extraction, classification, drafting, summarisation, support triage, and assistants that can call your own APIs under supervision.

Alongside that, the unglamorous automation that removes recurring manual work — approval flows, scheduled reports, and data moving between systems that were never designed to talk to each other. On the ASGA and SIDF SharePoint work, automating those paths cut manual operational steps by 40%.

How I work

I start from the task, not the model. We identify one workflow with a measurable cost, define what a correct output looks like, and build an evaluation you can run before anything touches production. Only then does the choice of model matter.

Prompts and tool definitions are versioned like code. Model calls go through a provider-agnostic layer, so moving between OpenAI, Claude and Gemini is a configuration change rather than a rewrite, and cost and latency are logged per call from day one.

What you get

• A working integration inside your product, not a demo in a notebook.
• Human review kept in the loop wherever a wrong answer would be expensive, with clear escalation paths.
• Guardrails: input validation, output schemas, retries, timeouts, and a hard cap on spend.
• Workflow automation in n8n or Zapier where a full build is not justified, documented so your team can extend it.
• An honest assessment of where an LLM is the wrong tool — often a query and a rule are cheaper and correct every single time.

The stack

OpenAI, Claude and Gemini APIs called from Node.js or Laravel services. n8n and Zapier for orchestration, webhooks for the connective tissue, and PostgreSQL and Redis for state, queues and idempotency.

A good fit if

Your team spends hours a week on work a machine could draft, or you have an AI feature on the roadmap and need someone who will ship it into production instead of prototyping it forever.