AI Agents & Assistants
AI agents that answer from your real data, not a generic script
Most AI agents fail for one reason: they're not actually grounded in the business's real data, so they guess. I build AI agents that read from real orders, documents and structured data before responding, and that hand off to a person the moment they're not confident.
Is this the right fit?
This is the right starting point if the problem is a repetitive question, enquiry or lookup that a person currently answers by hand — customer support, internal knowledge, lead qualification — and the business wants that handled without pretending an AI model can replace judgment it hasn't earned.
Problems this solves
- Staff repeatedly answer the same questions across scattered documents and tools
- Client or project knowledge lives in someone's head instead of a searchable system
- Existing chatbots give generic answers because they aren't grounded in real business data
- There's no safe way to let an AI model touch customer data without a human backstop
What I build
- A project scoped assistant that answers questions from indexed documents and structured data
- An AI agent that reads a customer's order and drafts an accurate reply, escalating anything it can't confidently answer
- An ongoing virtual assistant role covering operational and clinical support for a healthcare practice
- Retrieval augmented generation over a business's own content, so answers cite real source data instead of the model's own guess
Tools
Real proof, not a promise
AI agent and workflow automationShopify AI Customer Support Agent
An AI powered response agent built in n8n that handles routine Shopify enquiries automatically.
Read the case studyOngoing virtual assistant role, since April 2026Healthcare Operations and Clinical Support Virtual Assistant
An ongoing virtual assistant role covering healthcare operations and clinical support, rated 5 out of 5.
Read the case studyHow this works
Every engagement starts with how the process actually runs today, then moves through design, build, and testing against real data before handover. See the full five step process for the details.
Frequently asked
Will the AI agent ever make something up?
Every agent I build is designed to answer from real, retrieved data rather than the model's own memory, and to say it doesn't know rather than guess. The Shopify support agent, for example, checks a decision step before it ever replies, and forwards anything it isn't confident about to a person instead of improvising.
Do I need my own AI model or API key?
Usually yes — the agent runs on a model you control access and billing for (OpenAI is the most common), so you keep ownership of usage and cost, and I build the retrieval, logic and handoff around it.
Can an AI agent replace my support team?
It's built to remove the repetitive, answerable share of enquiries, not the judgment calls. The point of the escalation step is that a person still handles anything genuinely ambiguous — that's what makes it safe to run in production.
Have a process like this to automate?
Tell me what's slow or manual right now. I'll tell you what it takes to fix it.