Services · Prototypes

Prove it on your own data first.

I build one workflow on your own data, inside your own accounts, to production standard. If it passes, I harden that same build and put it into production instead of rebuilding it.

At a glance

usually, from start to decision
About 30 days
real cases from your own work
Your data
built where it would run in production
Your accounts
your decision at the end
Go / adjust / stop
  • One workflow at a time
  • Built on your own data
  • Built in your own accounts
  • Tested with the people who use it
  • About 30 days
  • Go, adjust or stop
  • Hardened and deployed, not rebuilt

What a prototype is

One workflow, built to production standard.

A prototype here is one workflow we pick together, built on your own data and inside your own accounts. I build it to production standard, because if it passes, that same build is the one I harden and put into production.

I don't build throwaway prototypes. If it works on real cases, you have the first piece of the production system. If it doesn't, you have spent weeks rather than committed to a platform.

One workflow, its inputs and a defined result
Scope
The success measure, agreed before the build
Measure
Agreed examples and acceptance criteria
Checks
Access you approve, and only that
Access

What I can prototype

Examples of what I can prototype.

Each one is a single workflow, simple to judge in about 30 days.

An agent that drafts replies from your documents

It answers from your own documents and drafts the reply for someone on your team to approve. It never sends on its own.

One reporting question, answered end to end

A question your organization asks every month and answers by hand. I build the retrieval and assembly for that single question completely, and if it works, answering the second question costs a fraction of what the first did.

An intake or quoting workflow

One intake or quoting flow built in a system you already run, with a review step so a person approves every quote or reply before it's sent.

An integration between two systems

Two systems you already use, such as your CRM and your accounting software, connected so no one retypes the same thing twice.

An internal tool

A small application for a step your software doesn't cover, such as a document format no software supports or an old system with no API.

How it's built

How a prototype gets built, in about 30 days.

Each step ends with something you can see, and I don't launch anything without your go-ahead.

  1. 01

    Pick one workflow

    We pick the workflow together, name the systems involved, and agree the success measure and the examples it has to get right.

    Scope and acceptance criteria agreed

  2. 02

    Build in your accounts

    I build it inside your own accounts, with the least access required, and only through access you approve.

    A working version on your data

  3. 03

    Test with your people

    The people who will use it test it against real cases from your own work. A working demonstration comes before any change to production systems.

    Test results and open risks

  4. 04

    Run a controlled slice

    Once you approve, it runs live on one slice of the work, such as one client group or one document type. I adjust the prompts and rules against real outcomes.

    Results from real work

The decision gate

Go, adjust or stop.

A planned checkpoint at the end, where you decide whether to go ahead, adjust or stop. The question is whether the working version met the success measure and the acceptance criteria we agreed in the first step.

At the end

What you get, and what I need from you.

What you get at the end

  • The working version, on your own data
  • The test results
  • The open risks, written down
  • A revised delivery plan for the next phase

What I need from you

  • A project owner on your side
  • Time with the people who do the work now
  • Access to the systems involved, approved by you
  • Sample documents and data

After the decision

If it passes, and if it falls short.

If it passes

I harden that same build for production, with role-scoped access, logging, versioning and rollback, and write the documentation your team needs to run it. It goes live once you approve in writing, and you get a fixed quote for the next phase before it starts.

If it falls short

You adjust it and test again, or you stop. Either way you have spent weeks rather than committed to a platform, and you can see in the test results exactly where it missed.

Tools

What prototypes are built with.

I build a prototype with the same tools the production version would use, on business accounts that belong to you.

AI models

  • Claude
  • ChatGPT
  • Copilot
  • Gemini

On business accounts, including OpenAI's or Anthropic's business platforms.

Automation

  • n8n
  • Make
  • Zapier
  • Custom code

Software

  • Next.js
  • React
  • TypeScript
  • Node.js
  • PostgreSQL
  • Redis

Hosting

  • AWS
  • Vercel
  • Cloudflare
  • Supabase

Or your own cloud account.

Security

  • OAuth
  • JWT
  • RLS policies
  • Rate limiting
  • Encryption

Your systems

  • Microsoft 365
  • Google Workspace
  • Your CRM

Questions

Before you start a prototype.

What does a prototype cost?

It's quoted in writing after a first conversation, once I know the workflow, the systems involved and the state of your data. You get a fixed quote before anything starts. On the pricing page I explain what makes a project cost more or less.

How long does a prototype take?

Usually about 30 days, ending in a go, adjust or stop decision. The exact timing depends on access, integrations and review, and I put the milestones in the written scope before work starts.

What happens to the prototype if we go ahead?

It becomes the production system. It was built to production standard in your accounts, so I harden it and put it into production instead of rebuilding it. I don't move anything into production without your written approval.

Do we need clean data first?

Usually not. Most useful systems read from your systems of record directly, and clean, consistent records can be used right away. If there are duplicates, old files or missing fields, those get fixed first, and I build that time into the scope.

Is our data used to train AI models?

No. I use your data only to do the work, I don't use it to train AI models, and I return or delete it when you ask in writing at the end of the engagement. The AI runs on business accounts, using frontier models from established providers under enterprise data terms.

Is the prototype built in our accounts or yours?

Yours. I build it in your Microsoft 365, Google Workspace, CRM or cloud account, with the least access required and only through access you approve. Once it's paid for, you own it.

The first step

Bring one workflow to a 30-minute call.

Tell me the task your team repeats most and which systems are involved. I'll tell you whether it would make a good first prototype and what I'd need from you to build it.