Skip to main content
The AI Cabinet

Services

Six rungs. One ladder. 

Start where you actually are. Most people begin with a Survey and end up on a Retainer, but nothing here requires you to buy the next thing.

The whole spec sheet. 

Fixed scope means a fixed price, quoted as a number rather than a range. Never hourly. Model and third-party costs sit on your own accounts and stay yours.

01

The Survey

Find what is worth automating

Two weeks. We sit with the people doing the work, map how it actually runs, and come back with a ranked list of what to automate, what it saves, and a fixed price to build the first one. If the honest answer is that a spreadsheet formula solves it, the report says that.

  • Ranked list, scored
  • A fixed price for build one
  • Fee credited in full
Open The Survey→

2 weeks

Fixed price, quoted up front

Credited in full against The Build

02

The Blueprint

Turn an idea into a spec

Four weeks on a single initiative. We take an idea that lives in your head and turn it into documentation a development team can execute from, including the architecture, the integration map, the threat model and a budget that survives contact with a CFO.

  • Architecture and integrations
  • Threat model
  • A budget that survives a CFO
Open The Blueprint→

4 weeks

Fixed price, quoted up front

Half credited against The Build

03

The Build

Live in 30 days, or free

One workflow, live in your own cloud and your own accounts, inside 30 days. With an eval suite, a runbook, a rollback path, alerting, and your team trained on it. Then we do the next one. This is the whole business: a thing that runs on a Tuesday when nobody is watching.

  • Your cloud, your repo
  • Evals, runbook, rollback
  • 30 days hypercare
Open The Build→

30 days to the first workflow in production

Fixed price, fixed scope

In production in 30 days, or you do not pay

04

The Hardening

Break it before someone else

One to two weeks on an AI system that is already live, or on the AI coding tools your developers are already using. We red team it, scope the permissions properly, and hand you a severity-rated report with the config changes already made. Run by Atul, whose 15 years are mostly in the part of this that goes wrong.

  • Red team, scored before and after
  • Least-privilege tool calls
  • Incident playbook
Open The Hardening→

1 to 2 weeks

Fixed price, quoted up front

05

The Retainer

Keep it alive when models move

Automation is not a project that finishes. Providers retire models, upstream APIs change their schema without telling you, and quality drifts quietly for weeks before anybody notices. The Retainer is the answer to the only question that matters six months after launch: is it still working, and who would know?

  • Migrations included
  • Drift watched on a schedule
  • One new workflow a quarter
Open The Retainer→

Month to month, 30 days notice either way

Monthly

30 days notice. No lock-in, rate held

06

The Pair

Both of us, on your backlog

Some companies do not need a discovery process. They need two people who can build, who have shipped in production before, and who can start on Monday. The Pair is us inside your team: your repo, your standup, your backlog, moving at the speed you actually work at.

  • Starts this week
  • Ships weekly
  • Documented as we go
Open The Pair→

Month to month, 30 days notice either way

Monthly

Both of us. One of us part time, quoted lower

How it is built

Agents get credentials. Act like it. 

Most of what goes wrong is a default nobody changed. Atul has fifteen years of production engineering and owns this on every engagement.

We hold no SOC 2 and no ISO 42001. A certificate from a two-person firm would be theatre. If procurement requires a certified vendor, we are the wrong choice and you will hear it on the first call.

Runtime
Your cloud. Your keys. Your accounts
Tool calls
Least privilege. Destructive actions gated
Training
Provider tiers that exclude your data
Models
Version pinned. Never whatever is newest
Quality
Eval suite on a schedule. You see the trend
Alerts
Failure, cost spikes, upstream schema drift
Before ship
Injection and exfiltration run as a suite
On exit
Code, prompts, evals, runbooks. All yours

Objections

The questions that kill deals. 

Where does our data go, and will it be used to train a model?

It runs in your cloud, on your API accounts, under your keys. We use business or enterprise tiers where the provider's terms exclude training on your data, and we put that in writing. During a build we see what a screen share shows us and nothing more. Anything that must never leave your tenancy is identified in the Survey and designed around, not discovered later.

Do you hold SOC 2 or ISO 42001?

No, and we will not imply otherwise. We are two people, and a certificate from a two-person firm would be theatre. What we do instead is answer your security questionnaire properly, build to least privilege by default, and have the person who did fifteen years of production engineering own that decision. If your procurement genuinely requires a certified vendor, we are the wrong choice and we will say so on the first call.

Who owns the code?

You do, from the first commit. Code, prompts, system instructions, orchestration logic, evaluation sets, documentation and runbooks, in your repository. There is no license to revoke, no platform to be locked into, and nothing to hand back if you stop working with us.

What happens when the model we built on is retired?

It will be. Providers deprecate model IDs constantly, sometimes with days of notice. We pin versions rather than tracking whatever is newest, and we keep an evaluation suite so that swapping a model is a measured decision rather than a hopeful one. On a retainer, migration is included. Without one, we will still tell you it is coming.

How do we know it still works in six months?

Because something is watching. The evaluation suite runs on a schedule and you get the trend, so slow quality drift shows up as a line going down rather than as a customer complaint. Alerting covers failure, cost spikes, and upstream schema changes, which is the failure nobody expects: the API did not break, it just started returning a slightly different shape.

You are two people. What if one of you is unavailable?

Both of us are across every engagement, and everything is documented while we build rather than afterwards, so either of us can step in. The code and the accounts are yours from day one, so the worst case is that you have a working system and full documentation and need a new builder, which is a normal problem rather than a crisis.

Can our own developers build from your plan?

Yes, and the documentation is written for that on purpose. The requirements, architecture and acceptance criteria are vendor-agnostic. Some clients take the Blueprint entirely in-house. Some take it in-house and come back for the one part nobody wanted to own.

What does it cost to run once it is live?

Model and third-party costs sit on your accounts, so you see them directly and they stay yours. For a single workflow it is usually a small monthly number rather than a meaningful line item, and we report cost per run as part of the build so it is never a surprise. We will give you the estimate before you commit, and we would rather over-estimate it.

What if the first project does not work?

The Build carries a shipping guarantee: in production in 30 days or you do not pay for the build. We do not guarantee results, because results depend on your data and your team, and anybody promising otherwise is selling you something. What we guarantee is the part that is in our control.

Do you work with companies outside your listed industries?

Regularly. We are two generalists, not a sector practice. The industries listed are where we have shipped, which shortens discovery, but the thing we are actually good at is learning an operation quickly and building for it. If we would be slow in your domain, we will tell you before you pay us.

How fast can you start?

We hold three active engagements at most, so it depends on what is running. A Survey can usually start within two weeks. If we cannot start when you need us, we will say that rather than sit on your enquiry.

Is there a lock-in or a minimum term?

No. Fixed-scope work is fixed scope. Retainers and embedded work are month to month with 30 days notice from either side, and the rate is held for the engagement.

Not sure which rung? 

Thirty minutes. No deck. Sometimes the answer is that you do not need us, and that is a short call.

Book thirty minutes

3 engagements at a time