Visual Network
Visibility Pilot
SEO, AI visibility and Google Business Profile audits, with reports of around 60 pages.
Read the case study on Visibility Pilot02 · Build something useful
From business problem to working AI application
You know the problem. You may not know which AI model to use, what belongs in the first version, how much it will cost to run, or whether AI is needed at all. That is the part I help with, before anyone writes code, and then I build it.
The kind of applications this suits
Focused tools that do one job well. For example:
- research and reporting tools that pull information together and write it up
- document checkers and summarisers
- lead research and qualification
- content support with a person approving the output
- internal dashboards for information that currently lives in spreadsheets
- small customer-facing tools, such as quote builders, checkers and calculators
- multilingual tools, in English and French
It is not the right service for large enterprise systems, apps that must handle medical or financial decisions on their own, or a generic chatbot bolted onto a website.
How a build goes
Define
The problem, the people who will use the tool, what they do today, and what a good result looks like.
Plan
The features for the first version, the data it needs, what must stay private, and roughly what it will cost to run each month.
Design
How it looks and works on a phone and a laptop, including the empty screens, the error messages, and where a person checks what the AI produced.
Prototype
The smallest version that tests the central idea. A working link, early.
Test
Accuracy, speed, running cost and failure cases, with real users where possible.
Refine
Then an honest decision: a stronger prototype, a first live version, a larger build, or stop.
Prototype or finished application? I will always say which
| What you get | What it is for |
|---|---|
| Product concept | A written description and sketches. Tests whether the idea makes sense. |
| Design prototype | Screens you can click through. Tests whether people understand it. |
| Interactive prototype | A working link with the AI connected. Tests whether the idea works. |
| Proof of concept | One hard technical question answered, for example "can it read these documents reliably?" |
| First live version | Real users, proper sign-in, error handling and hosting. The smallest version you can run a business on. |
| Production application | Live, maintained and monitored, with the security and backups that real data needs. |
A prototype is not built to carry real customers or sensitive data. If that is where the project is heading, the security work is planned and priced as part of the build, not added at the end.
Two technical routes, in plain words
A tool that runs in the browser. No accounts, no database, nothing to maintain. Right for calculators, checkers and quote builders. Cheap to host and fast.
An application with a server. Needed as soon as people sign in, save their work, pay you, or use an AI model without your key being visible to the public. More to build, more to look after, and the right choice more often than people expect.
I will tell you which one you need, and why, before you spend anything.
Security and data, taken seriously
The questions that matter are settled in the planning stage, not after launch: who can sign in and what they can see, where data is stored and for how long, which AI provider processes it, how API keys are kept off the public side, what happens when the AI gets it wrong, and what stops someone running up your bill. HTTPS alone does not make an app secure, and I will not say it does.
Apps I have built and run
Lead Finder
Finds local businesses with no website or a weak Google profile and drafts a first message for a person to review.
Read the case study on Lead FinderShowroom
Prepares outreach: checks the evidence, drafts a one-page proof and a message, and nothing leaves until I have approved every claim.
Banger Forge
Writes a complete Suno music brief in four steps, with a choice of OpenAI, Anthropic Claude or Google Gemini and automatic fallback if one is down.
Every application here is one of my own, built and run day to day. I would rather show you those than invent client logos.
Already built and available to licence: Lead Finder, on a member account for agencies and freelancers, rather than as a new build. Other apps may follow.
Questions
Do I need a full technical specification?
No. A clear description of the problem and who has it is enough to start. Working out the rest is the Define and Plan stages.
How much does an app cost?
It depends on the route and the scope, so it is quoted after a first conversation. A browser-based tool is a very different job from an application with accounts and payments. You will get a figure, a timescale and the monthly running cost before you commit.
How long does it take?
A focused browser-based tool is usually days rather than weeks. An application with sign-in and payments takes longer, because authentication, billing and hosting are jobs of their own. Either way, you get a working link early.
Which AI do you build with?
I build with Claude, ChatGPT and Google AI Studio directly, not with one-click app builders, because their output is hard to fix when it breaks. Inside your app, the model is chosen for the job: cost, speed, accuracy and where the data is processed.
Can you work with my existing developer or agency?
Yes. I can scope and prototype the idea, then hand it over with notes, or review something they have built.
Can you improve a prototype someone else made?
Yes. I will start by telling you honestly whether it is worth improving or quicker to rebuild.
What if AI is not the answer?
Then you will be told, and pointed at the simpler option. Sometimes a spreadsheet, a form or a £30 subscription does the job.