Skip to content
AlanVaneAI & Software

I build AI agents
and automation that
businesses actually run.

I'm Alan, with 20+ years of full-stack experience, focused on Cloudflare full-stack architecture, AI applications, and cross-platform development. I build AI agents, workflow automation, and custom web and mobile apps for clients worldwide.

Full-stack engineering, remote delivery, a one-person company, serving clients worldwide.

TypeScript, Go, and Flutter. I work solo, which means you talk to the person who writes the code, not an account manager.

agent-run.log
$agent run triage --inbox enquiries
→read 9 new enquiries since yesterday
→sorted each one by service and urgency
→drafted a reply for the 6 that were clear
!held 3 back — budget and timeline too vague to quote
→left those for you with a note on what's missing
✓nothing left waiting on you
Pattern: command → steps → human escalation → completion

"Can they do it?"

20+

years shipping software

"How long? Is it worth it?"

1 week

to a working prototype on your own data

"Will I be handed off?"

0

account managers between you and the code

"Will I be locked in?"

100%

of the code delivered into your repository

What I build

Custom AI solutions and software — all of it code I write, test, and hand over. If an off-the-shelf tool solves your problem for the price of a subscription, I'll tell you that instead.

Cloudflare full-stack architecture & migration

Run Next.js / Nuxt steadily on Cloudflare and say goodbye to a runaway Vercel bill — costs drop significantly.

  • Migrate Next.js / Nuxt to Cloudflare Workers for significantly lower costs
  • R2 object storage, D1 / Hyperdrive data layer integration
  • Edge caching, KMS, image processing and other production practices

AI agents

Agents that read, decide, and act inside your systems, not chatbots that answer FAQs. Give one a job that needs judgement — processing documents, sorting enquiries, research, monitoring — and it works through the task and escalates whatever it isn't sure about.

  • Read what came in
  • Decide what the task needs
  • Act, or escalate to a human
Learn more →

Business automation

AI automation that removes the repetitive work between the tools you already use. Quotes, invoices, data entry, reporting, document extraction. These are the jobs someone on your team currently does by re-typing things from one screen into another.

  • The invoice arrives
  • Fields are extracted
  • Written to your accounting system
Learn more →

Custom software

Internal tools, web apps, and mobile apps. TypeScript, React, Go, and Flutter. Fixed scope, code in your repository, documented and handed over. Built to be owned, not rented back to you.

  • Scope fixed up front
  • Weekly demos during the build
  • Delivered into your own repository

Cross-platform & mobile apps

One React Native / Expo codebase, shipped reliably to iOS and Android.

  • React Native / Expo app development and store publishing
  • Native build troubleshooting (e.g. 16KB page size)

US company registration & overseas payments

A one-time $589 package covers US company registration, EIN, bank account opening assistance and Stripe payment setup.

  • US LLC registration and EIN application steps, walked through
  • Help preparing US business bank account and Stripe application materials
  • Connecting the company entity, payment accounts and product billing systems

Not sure which of these you need? A free discovery call is where we work that out first.

Built, and written up

I publish how these work: the architecture, the code, and the parts that didn't work first time. You can check the reasoning yourself before you ever email me.

Cloud Code (Cloudflare + OpenCode)

A browser-based AI coding workspace. One Cloudflare Worker fronts a stateful container that runs OpenCode, deployable to your own Cloudflare account.

The idea was simple: a browser-based AI coding workspace shouldn't be a black box — you should be able to run it yourself, understand it, and extend it. I built it out from there: OpenCode v2 support, container image upgrades, CI smoke tests, and documentation for deployment, Cloudflare Access, and local testing.

TypeScriptCloudflare WorkersCloudflare ContainersOpenCodeDocker

Self-hosted, not a hosted product. Supports HTTP Basic Auth; Cloudflare Access is recommended for production.

How it works

Most decisions that keep getting deferred are deferred for the same reason: what you need to make them lives in several places, and nobody has half a day to assemble it. An agent assembles it on a schedule, works out what it means, and either acts or puts a recommendation in front of you. The shape holds wherever a decision depends on things your own systems can't see.

What it looks at

Your systems' history

The history sitting in the tools you already own: order and margin history, job and quote records, invoice and supplier patterns. Not just what is biggest, but what is accelerating, what is quietly dying, and what only moves when it is discounted.

What they can't see

The signal from outside: what competitors are doing and pricing, where public demand and search interest are heading, and the context that moves demand — season, weather, upcoming events, regulation. Obvious in hindsight, and almost never in the spreadsheet.

  1. 1
    your systems — history, records, numbers
  2. 2
    outside signal — competitors, trends, season
  3. 3
    work out what it means — real shift or noise
  4. 4
    decide what to change — stock, price, copy
  5. 5
    act or queue it — your rules decide which

The path a scheduled run takes, from your own records and outside signal through to a decision and an action. Steps in order: your systems, then outside signal, then work out what it means, then decide what to change, then act or queue it.

From recommendation to action

What comes back is a decision with an action attached, not a dashboard. A dashboard is more work; this is less. You set what the agent is allowed to do on its own, and what waits for you. Most people start with everything queued for approval, watch it for a few weeks, then hand over the low-risk actions once they have seen it work. Anything irreversible always goes to a human. What an action is depends on the work:

  • Adjusting a price, or flagging stock that is tying up capital.
  • Drafting the follow-up or the quote, so it only needs checking.
  • Rewriting a listing or a description to match what people are searching for.

A model for each kind of step

The high-volume mechanical steps — reading a page, pulling the fields out of it — run on small fast models. The handful of genuine judgement calls run on a frontier model. Routing each step to the model that suits it is most of what keeps an AI agent affordable to run without making it worse where it matters.

When it's just automation

Custom AI solutions get built one at a time. If your version has no real decision in it and is mostly moving information between systems, that is workflow automation instead, and cheaper to build.

Hi, I'm Alan

Remote · serving clients worldwide

I've been writing software professionally for more than twenty years.

I started out building websites in 2006, back when it was plain HTML. Those years taught me that shipping is mostly the unglamorous part: performance, edge cases, and the last ten percent that takes as long as the first ninety.

From there I moved into full-stack development, and that has been most of the work since. Mostly TypeScript, React, and Cloudflare, mostly SaaS products and the teams behind them, where slow or broken software turns straight into lost customers and lost revenue. Alongside that I built my own products — a form backend, an online image processor, small tools I use every day — which is still what I reach for when a job needs something small and sharp instead of a platform.

I started putting AI into my daily work in early 2024. The first thing I built with it was a browser automation tool: from a search keyword it pulled back YouTube videos about a certain brand's product — titles, view counts, likes — and exported the results as an Excel file for the operations team.

It changed how AI coding feels to me: the gap between an idea and something that runs on real data disappeared. The standard stayed what it always was — AI-written code gets read, tested, and shipped by me.

These days I work as a one-person company: requirements, architecture, coding, launch, operations — all of it mine, with no middle layer and no sales pitch. Remote and async is how I work best, moving by milestones, quoting per project. Working solo is deliberate: projects get scheduled so I never have to subcontract — which means the person you talk to on the first call is the person who writes the code, tests it, and hands it over.

These are the areas people come to me for most:

  • Cloudflare full-stack architecture and migration — running Next.js / Nuxt properly on Workers, wiring up R2, D1 and Hyperdrive, and getting teams off runaway Vercel bills.
  • AI agent products — from OpenAI / Claude integration to Stripe + Supabase usage metering and subscriptions, getting an AI product that actually charges money working end to end.
  • Cross-platform and mobile apps — React Native / Expo development, shipped to the App Store and Google Play.

Outside work I'm a runner, and I love travelling with my family.

If you have a product you want to build, or a stubborn engineering problem, email me and let's talk.

What I build with

TypeScriptReactNext.jsGoNode.jsPostgreSQLCloudflare WorkersDockerOpenCodeClaudeOpenAILocal LLMs

How I work

Direct access

You talk to the person writing the code. There is no account manager in the middle and nothing gets handed to an offshore team.

Remote collaboration

Comfortable with async communication and cross-timezone collaboration, moving projects forward by milestones.

20+ years of full-stack experience

From frontend to infrastructure, I carry complete products on my own.

Read, tested, shipped

AI speeds things up considerably, but the standard stays what it always was — AI-written code gets read, tested, and shipped by me.

Quoted per project

Estimated by project scope, clear and transparent, no long-term lock-in.

Fixed scope

A written scope and a price before anything starts. If the work grows, we agree that separately rather than watching an hourly rate run.

Code you own

It lives in your repository, under your account, documented, with a proper handover. Nothing is a black box you can only maintain by paying me.

No retainer lock-in

Ongoing support is available if you want it. It is never a condition of getting the work built in the first place.

Questions

Straight answers about agents, automation, and what building this actually involves.

A chatbot answers questions. An agent does work. It has access to tools: your inbox, your database, an API. Then it loops: it takes an action, looks at what came back, and decides what to do next, until the job is done or it decides a human should look at it. If what you need is "answer questions from our FAQ", that's a chatbot and it's cheap. If you need "read this, work out what it is, and put it in the right place", that's an agent.

Yes. It lives in your repository, under your account, documented, with a proper handover. You can maintain it yourself or hire someone else to. There is no licence that expires and nothing that stops working if you stop paying me.

It depends almost entirely on two things: how many systems have to talk to each other, and how messy the input is. One well-defined workflow between two systems that already have APIs is a small, fixed-price project. The same workflow where the input is scanned PDFs from dozens of different suppliers is a much bigger one. I quote a fixed price per project after a short discovery, so you know the number before you commit to the build. Running costs after that are usually modest, being API usage plus hosting.

For a single workflow, usually two to four weeks. About a week of that is a working prototype you can try on your own data before the rest gets built. That is deliberate, because it is much cheaper to find out early that the idea needs changing. Larger builds get staged so you see something working at the end of each stage rather than only at the end.

Usually, yes. Most modern SaaS have APIs, and that is the straightforward case. Older or custom internal systems are normally still workable through a database connection, a file drop, or whatever integration surface exists. It is the part most worth checking before anything else, so it is the first thing I look at.

A lot of work runs through Claude or OpenAI, and for most business data that is fine. If your data cannot leave the country or the building, I can assess whether a local model makes sense for your case. That is a genuine trade-off rather than a sales line: local models are cheap to run and completely private, and they are less capable than the frontier ones. Which way to go depends on the task, and I will tell you which one I think it is.

No. Search for an AI agency and you will find plenty of them, and for a large programme of work one of them may well be the right answer. I am one developer. For a single agent or one automated workflow that usually means a decision in an hour rather than a week, no account layer to pay for, and the person you brief is the person writing the code. If your project genuinely needs a team of ten and a 24/7 SLA, I will tell you so and point you at an agency.

Most of the work is remote, and clients are spread across time zones. For this kind of work the timezone matters considerably more than the postcode. My main working hours are 06:00–21:00 (UTC+8).

Start a project

Tell me what you're trying to do and I'll tell you whether it's worth building. If it isn't, I'll tell you what I'd use instead. The first conversation doesn't cost anything.

  1. You get a reply within one business day.
  2. We have one conversation about what you're trying to do. Call or email, whichever you prefer.
  3. If it's worth building, you get a written scope and a fixed price before anything starts.
Project Type (select all that apply)
06:00–21:00 (UTC+8) · reply in 1 business day