Artificial Intelligence, When It Genuinely Helps

Where AI actually helps

AI powered assistants trained on your own product, service, or documentation

When customers or staff keep asking the same questions and the answers already exist somewhere, scattered across a manual, a knowledge base, or a support inbox, an assistant trained on that material can answer instantly and correctly instead of someone digging through documents or waiting on a reply. This works well when the underlying information is already there and just needs a faster way in, not when the goal is to sound impressive on a landing page.

Automation of repetitive manual tasks

Data entry, quoting, and reporting are the tasks I see eating the most hours for the least reward. A person retyping numbers from a photo, an invoice, or a form into a spreadsheet every day is a strong candidate for AI, because the task is repetitive, well defined, and error prone precisely because it is boring. Automating that step usually pays for itself quickly, both in time saved and in mistakes caught.

Natural language interfaces for internal tools and dashboards

Some internal tools hold a lot of useful data behind filters, dropdowns, and report builders that only the person who built them really knows how to use. A natural language layer on top lets someone ask a plain question and get a straight answer, without learning the tool underneath. I use this when a team's internal software has grown more complex than the people using it day to day actually need it to be.

Integration of AI capabilities into existing websites, apps, or internal systems

Most of the time AI does not need its own standalone product. It needs to slot quietly into something that already exists: a website that can answer a visitor's specific question, an internal system that flags something unusual before a human even looks at it, an app that reads a document and pulls out the fields someone would otherwise type by hand. The goal is always a system that works better, not a new interface to learn.

Real world data feeding AI decisions on sites and in factories

A lot of useful decisions happen too late because the person who could act on the data is not looking at it at the right moment. I connect AI to sensors and real time data streams already present on a site or a factory floor, temperature, throughput, equipment status, occupancy, whatever is being monitored, so the system can flag a developing problem, spot an optimisation opportunity, or support a decision before a human would have caught it by checking a dashboard manually. This sits on top of monitoring that already exists, it does not replace the people watching the site, it gives them earlier and clearer signal.

Proven work

A food manufacturing plant was recording every incoming pallet by hand under real time pressure, weight, temperature, farm data, all written on paper while a forklift kept moving. I built a tool that uses Gemini 2.5 Flash to read photos of the supplier sheet, scale, and thermometer, extract the relevant data, and cross verify it automatically, flagging mismatches without slowing the operator down. It took one week to build, and it runs for under $0.15 a month in API calls. It has already caught real data mismatches that would otherwise have gone unnoticed until much later.

This is the clearest example of how I actually use AI in a project: a specific, repetitive, well defined task, replaced with something faster and more reliable, at a running cost close to zero.

How I think about AI in a project

I do not start a project by deciding to use AI and then looking for a way to justify it. I start by understanding the problem, the same way I would for any project, and AI comes into the conversation only if it is genuinely the best way to solve part of it. On the home page I describe AI as one tool among the ones I use, applied where it earns its place, and that is not a marketing line, it is how I actually scope work.

In practice, this means that if a simpler, non AI solution solves your problem better or more cheaply, such as a well designed form, a basic automation rule, or a clean piece of standard logic, that is what I will recommend instead. I would rather deliver something boring that works reliably and costs little to run than something impressive sounding that adds complexity, cost, or unpredictability without a real benefit to the person using it. Being genuinely polyvalent means I have no incentive to push every project toward the same tool, because AI is not the only thing I offer.

When AI is the right call, I still keep it scoped to the specific task it is good at, extracting data, answering a well bounded question, automating a defined step, rather than wiring it into everything simply because it is available. That is what keeps these systems fast, cheap to run, and predictable, instead of expensive and hard to trust.

What's included

• AI powered assistants trained on your own product, service, or documentation

  An assistant that actually knows your material and answers correctly, built from the documentation, catalogue, or knowledge base you already have.

• Automation of repetitive manual tasks such as data entry, quoting, or reporting

  Well defined, repetitive work handed to a system that does it faster and more consistently, freeing up the person who currently does it by hand.

• Natural language interfaces for internal tools and dashboards

  A plain language layer over software that has grown more complex than the people using it day to day actually need it to be.

• Integration of AI capabilities into existing websites, apps, or internal systems

  AI slotted into what you already have, rather than a new standalone product nobody asked for.

• Real world data feeding AI decisions on sites and in factories

  Sensors and live data from a site or plant connected to AI that flags problems early, spots optimisation opportunities, and supports decisions for the humans already monitoring the operation.

From $1,800 USD

Common questions

Will you just bolt on a chatbot?

No. A chatbot bolted onto a website that does not need one is decoration, not a solution, and I will tell you if that is what is being asked for. AI gets added where it removes real friction from a real task, not as a default feature or a way to make a project sound more current. If a chatbot genuinely is the right answer for your case, I will build one, but it has to earn that place first.

What AI models do you use?

It depends entirely on the task. For structured extraction work, reading a photo or a document and pulling out specific fields, a fast, cheap model like Gemini 2.5 Flash is usually the right fit, which is what I used for the goods reception case study above. For tasks where reasoning quality matters more than speed or cost, I use a more capable model instead. I choose the model based on what the task actually needs, not on habit or whichever one is trending.

Is this expensive to run long term?

For a well scoped AI task, usually not. The goods reception tool referenced above runs for under $0.15 a month in API calls, and that is typical of tasks that are specific and well defined rather than open ended. Running costs typically land somewhere between a few cents and a few dollars a month. I will give you a realistic running cost estimate before we start, so there are no surprises once the project is live.

Do you handle data privacy and security for this?

Requirements here vary a lot by industry and by the kind of data involved, so I discuss and scope data handling explicitly as part of every AI project rather than applying a single default approach. That includes what data is sent where, what gets stored, and what provider is involved. If your industry has specific requirements, tell me upfront and I will design around them from the start, not retrofit them afterward.

Start a project

Send me a short message describing the task or process you think AI could genuinely help with. I read every message personally and reply within one business day, and I will tell you honestly if AI is not actually the right tool for it.

Start a project