Your clients are asking for AI.
We build it under your name.
White-label AI engineering for agencies that have the client relationship but not the AI team. Your brand on the delivery, our engineers on the build, and an eval harness that proves the thing works before it ever reaches your client.
The situation
The request arrived. The team to build it did not.
Every agency is getting the same brief this year, and most are answering it one of three ways: turn it down and watch a retained client go shopping, hand it to a generalist and hope, or hire an AI engineer for a pipeline that is not yet predictable enough to justify the salary.
Subcontracting is the fourth answer, and it usually fails for one reason: you cannot verify the work. AI is exactly the discipline where “it looks like it works” and “it works” are different things, and the gap shows up in front of your client rather than in your QA. That is the problem we build around.
How it works
You keep the relationship.
We keep the pager.
- 01
You bring the brief
From your client, in whatever shape it arrived — a scope doc, a call recording, or a one-line ask you have not scoped yet. We do the technical scoping.
- 02
We scope and quote to you
Fixed scope, fixed price, stated timeline. You mark it up however you like — we never see or ask about your client-facing number.
- 03
We build under your name
Your repo or ours, your stack conventions, your project manager in the loop. We can stay entirely behind you, or join client calls as your team — your call, per project.
- 04
You hand over, we stay reachable
Documentation written for your team, not for us. A support window after handover so you are never the one who cannot answer a question.
What we take on
The AI half of your project.
We are a specialist subcontractor, not general capacity. That is deliberate — it is why we can put a number on the quality.
RAG agents and chatbots
Retrieval over your client's real documents, catalog or internal data. Grounded, so it cannot invent facts in front of their customers.
LLM features in existing software
The AI surface your client asked for, built into the app you already maintain — without rewriting the parts that work.
Workflow automation
Agentic automation that removes manual steps for good, wired into the tools their operations already run on.
Evals and reliability rescue
An AI build already shipped and misbehaving — yours or someone else's. We measure it, rank the failures, and fix them in order.
Terms, up front
What we sign before we start.
Mutual NDA before scoping
Signed before we see anything of your client's. Standard, not negotiated per project.
Your client stays your client
Non-solicitation in writing. We do not approach them during the engagement or after it.
Fully white-label by default
No Caffinix in the repo, the docs, the commits or the deliverable unless you ask for it.
Fixed scope, fixed price
You quote your client with a number that will not move under you mid-build.
Questions
What agencies ask us first.
Yes, by default. Unless you ask otherwise, nothing we deliver carries our name — not the repo, the documentation, the commit authors or the handover deck. We can stay entirely behind you, or sit on client calls as part of your team. It is decided per project, by you.
No, and it is in writing. Non-solicitation is part of the standard agreement, during the engagement and after it. Your client relationship is the thing you are actually protecting when you subcontract, and we treat it that way.
We are founder-led with an on-demand senior bench — engineers we have worked with before, staffed per project rather than sat on payroll. That means we will tell you honestly what we can staff and by when, instead of promising a team we would then have to go and find. If a project needs more than we can cover, we say so before you quote it.
Scoping usually starts within a week. An eval audit can run in one to two weeks; a full RAG or agent build is typically four to eight weeks. We give you a date you can put in front of your client, and we tell you early if it is going to move.
That is the point of the eval harness. We agree a quality threshold with you before the build starts, measure against it, and do not call the work done until it clears. If it turns out the approach cannot clear the bar, you find out during the build with a measurement in hand — not after your client has shown it to their customers.
Yes, when it is adjacent to the AI work — the APIs, data pipelines, and infrastructure an AI feature needs to actually run. We are not the right subcontractor for general WordPress, Shopify or marketing-site overflow, and we will point you elsewhere rather than take it.
We are in Dhaka (UTC+6), which gives full-day overlap with the UK, Europe, the Middle East, Australia and Asia, and morning overlap with the US East Coast. We commit to a response window in the agreement rather than asking you to guess.
Have a brief you cannot staff?
Send it over. We will tell you what it actually takes to build, what it would cost you, and whether it is worth quoting at all — before either of us signs anything.