I build AI workflows for teams who are done with slide decks
Most AI consultancies will sell you a strategy document and a roadmap. I’ve been on the receiving end of those, and the document usually ends up in a shared drive somewhere while the team carries on doing everything manually. So I do it differently. I find the bottlenecks in how your team actually works, I build the automations that clear them, and I train your people to keep it all running after I’m gone.
I’ve spent 20 years in search and digital strategy (I founded Builtvisible, which was later acquired), and more recently I’ve been building AI tooling full-time. I’ve published 16 open-source MCP servers, I run local LLMs on my own multi-GPU hardware, and I use every tool I recommend to clients on my own projects before I suggest it to anyone else. If something doesn’t work in practice, I don’t sell it.
Book a free 30-minute callWhat I do
Workflow Audit
I spend two weeks looking at how your team actually works (not how they think they work, which is always different), and I hand you a prioritised list of what’s worth automating, what isn’t, and what order to tackle it in. Most teams find the audit pays for itself within a month because the wasted time turns out to be far more expensive than anyone realised.
AI Automation Build
Once we know what to build, I build it. Document processing, research pipelines, content production, data transformation, report generation, whatever the audit identified as highest-value. These are multi-step systems using n8n, Claude, Python, Cloudflare Workers, and custom MCP servers, and they run on your infrastructure so you own the whole thing.
Custom MCP Servers
Your CRM, your ERP, your internal databases, your weird legacy system that nobody wants to touch but everyone depends on. These all need custom connectors before an AI assistant can use them properly. I’ve built 16 production MCP servers (all open-source, all on npm), and I build yours the same way. You own it, no licence fees, no lock-in.
Team Training
Hands-on workshops where your team builds real workflows on their own laptops using your actual data. By the end of the day everyone has Claude Desktop configured, MCP servers installed, and at least one working automation they built themselves. I don’t do slides and theory because in my experience people forget those within a week.
Fractional AI Lead
A monthly retainer where I attend your planning meetings, guide tool selection, oversee implementation, and gradually upskill your team so they don’t need me anymore. The goal is explicitly to make myself unnecessary within 3 to 6 months, because if you still need me after that then I haven’t done my job properly.
AI Content & AEO Strategy
I’ve been working in search since 2004, and I built the content pipeline and the voice analysis tooling that produces stg-houtiniwp-houtini.kinsta.cloud. If you need a content strategy that’s designed for both traditional search and AI-generated answers (answer engine optimisation), this is the bit where the 20 years of experience is probably most directly useful.
How this usually works
We have a conversation
A free 30-minute call where you tell me what’s slow, what’s frustrating, or what you’re trying to figure out. I’ll be honest about whether AI is the right solution, and if it’s not then I’d rather tell you that now than take your money and discover it later.
I send you a proposal
A plain-English document explaining what I’d build, roughly how long it would take, and what it costs. I don’t do 40-page proposals because in my experience nobody reads them, and the ones that do get read tend to raise more questions than they answer.
I build it and hand it over
Most engagements take somewhere between 2 and 6 weeks. I build it, you test it, I document it, you own it. There’s a 30-day support window after delivery because things always come up once a system meets the real world, and I’d rather deal with those early than have you stuck with something that doesn’t quite work.
Questions I get asked
Do we need technical people on our team?
No. The whole point of what I build is that non-technical people can use it. I design the systems, I train your team on how to operate them, and I make sure the documentation is good enough that someone who wasn’t in the training session can still figure it out. If something goes wrong, you call me rather than trying to debug it yourself.
Can you work with our existing tools?
Almost certainly. I’ve built integrations with Google Workspace, Slack, Salesforce, HubSpot, WordPress, Shopify, Brevo, Amazon product feeds, financial data APIs, Google Search Console, and quite a few proprietary systems I’d never heard of before the client introduced me to them. I’ve also connected to spreadsheets that probably should have been databases years ago, which is more common than anyone likes to admit. If it has an API or a file export, I can connect to it.
What about data privacy and security?
This is something I care about independently of client work. I run local LLMs on my own multi-GPU hardware (a Threadripper workstation with six NVIDIA cards) precisely because some data shouldn’t leave the building. If your work involves sensitive information, I can build with local models like Ollama and LM Studio that keep everything on your infrastructure, with nothing going to a cloud API at all. I wrote a complete guide to setting up LM Studio that explains how this works in practice.
How is this different from just buying Copilot or ChatGPT Enterprise?
Those are general-purpose tools, and they’re useful for general-purpose things like drafting emails and summarising documents. But they don’t know your processes, your data structures, or your terminology, and they can’t connect to your internal systems without custom integration work. What I build is specific to how your team actually operates. The difference is a bit like buying Microsoft Excel versus hiring someone to build you a financial model that fits your actual business.
What happens when something breaks after you leave?
Every engagement includes a 30-day support window, and I document everything thoroughly enough that your team (or a future developer) can maintain it without me. But honestly, the things I build tend to be pretty stable because I use the same infrastructure patterns for every client, and those patterns have been running on my own systems for months before I deploy them anywhere else. If something does need attention after the support window, I’m not going to ignore your email.
What’s your tech stack?
TypeScript and Python for most things, n8n for workflow orchestration, Cloudflare Workers and D1 for edge deployment, SQLite for local data, Crawlee for web scraping, Claude and Gemini APIs for reasoning tasks, and local LLMs via LM Studio and Ollama for privacy-sensitive work. I’ve also built integrations with the Amazon Product Advertising API, Financial Modeling Prep, Google Knowledge Graph, Brevo for email automation, and various recruitment platform APIs. The honest answer is that the stack depends entirely on what the client needs, and I pick tools based on what I’ve already proven works rather than whatever’s trending on Hacker News this week.
Not sure where to start?
Book a free 30-minute call. Tell me what’s eating your team’s time and I’ll give you an honest assessment of whether AI can help, and if so, what the first step would be. No pitch, no slides, no follow-up sequence.
Book a call