Do you build production systems or just advise?

We build. Engagements produce running code in your repositories, under your review process, with your team involved throughout — the goal is that they own it fluently when we leave. We are not interested in becoming a dependency you cannot remove.

Which models and frameworks do you work with?

We are deliberately model-agnostic and design for substitution, because the frontier moves faster than most procurement cycles. In practice that means an abstraction at the inference boundary, an evaluation set that makes swapping measurable, and no framework in the critical path that we would not be willing to fork.

Why does the security practice touch the build work?

Because retrofitting authorization onto a retrieval pipeline is far more expensive than designing it in, and because the failure modes we spend our time exploiting are cheapest to prevent at the architecture stage. Build engagements get security review as part of the work, not as an upsell.

Can you take over an existing system?

Yes, and it is a common starting point — usually a prototype that reached production faster than intended. The first phase is an honest assessment of what is salvageable, which occasionally concludes that a component should be replaced rather than repaired. We will say so plainly and cost both paths.

What size team do you field?

Small and senior. Boutique is a deliberate constraint, not a stage we are trying to grow out of — it is why the same people who scope an engagement are the ones who deliver it. It also means we turn work down when we are full, and will tell you that rather than staffing it thin.

Build it

AI Engineering

Production AI systems built by people who spend the rest of their time attacking them. Evaluation harnesses, retrieval pipelines, agent orchestration, and the serving infrastructure underneath — designed so the security review and the invoice both hold up later.

Scope an engagement

Engagements from $16,000. Or start with the Exposure Review at $4,500.

Pricing status: confirmed. PinkCloud quotes from these figures.

Prices exclude any applicable tax.

What an engagement looks like

Scoped as fixed-window work with a written deliverable. No retainers you cannot exit, no findings held back for a follow-on sale.

Evaluation harness

2–4 weeks · build

The thing most teams skip and later wish they had not. A test suite for behaviour: task-level success criteria, regression cases drawn from real failures, and adversarial cases from the security practice. Run in CI, with results that move when the model, the prompt, or the retrieval corpus changes — so a swap can be evaluated instead of guessed at.

Retrieval pipeline

4–8 weeks · build

Ingestion, chunking, embedding, and ranking built with authorization as a first-class concern rather than a later retrofit. Per-tenant isolation enforced at query time, access changes in the source system reflected in the index, and evaluation of retrieval quality independent of generation quality — because when answers are wrong, it matters which half failed.

Agent orchestration

5–10 weeks · build

Systems where the model calls tools and the calls have consequences. Least-privilege tool design, deterministic guardrails outside the model rather than inside the prompt, idempotency and replay safety on every side-effecting action, and observability that captures the reasoning trace alongside the API calls it produced.

Inference infrastructure

3–6 weeks · build or review

Serving architecture with unit economics as a design constraint: batching strategy, caching at the semantic and exact layers, quantization trade-offs measured against your evaluation set rather than a public benchmark, and routing between model tiers so the expensive one handles only what needs it.

Principle

If you cannot measure the behaviour, you are not engineering it.

A great deal of AI work is shipped on vibes: someone tries a dozen prompts, the outputs look reasonable, and it goes live. Then the model version changes, or the corpus grows, or a user phrases something unusually, and there is no way to tell whether the system got worse because there was never a measurement.

Every build engagement starts with the evaluation harness — not because it is glamorous, but because it is the only thing that makes every subsequent decision checkable. It is also what turns a security finding into a regression test instead of a memo.

Design constraint
Guardrails outside the model
Anything enforced only by instructions in the prompt is a request, not a control.
Design constraint
Authorization at query time
The index enforces the caller’s permissions, not the ingestion job’s.
Design constraint
Side effects are idempotent
An agent that retries must not double-charge, double-send, or double-delete.
Design constraint
Substitutable inference
Model choice is a configuration decision with a measurable answer, not a rewrite.

Order

AI Engineering Assessment

Two weeks on a production AI system or its design: an honest assessment of what is salvageable with both paths costed, and the evaluation harness that makes every later decision checkable.

Two weeks at the published floor for the AI engineering practice. The floor covers up to 3 distinct systems, models or accounts; beyond that, each further block of up to 3 systems adds 10 percent, capped at 30 percent, and more than 12 systems needs a scoping call.

Two weeks buys the practice's shortest shapes, chosen at the scoping call: an architecture review of a system or a design, an honest assessment of what is salvageable in an existing system with both paths costed, or the Evaluation harness (the site scopes it at 2–4 weeks, build) at its two-week end. The harness is the shape with a published week range; the architecture review and the salvageable assessment are sized against the two-week floor at that call, and where a system is larger than two weeks of senior time can honestly cover we say so then and quote the work as an engagement rather than taking the fixed fee. Taking over a prototype that reached production faster than intended is a common starting point; the assessment occasionally concludes that a component should be replaced rather than repaired, and we say so plainly and cost both paths.

The system is read against the design constraints the practice states: guardrails outside the model rather than instructions inside the prompt, authorization enforced at query time by the index rather than by the ingestion job, side effects that are idempotent so a retrying agent cannot double-charge, double-send or double-delete, and substitutable inference, where model choice is a configuration decision with a measurable answer rather than a rewrite.

assessment

Two weeks on a production AI system or its design: an honest assessment of what is salvageable with both paths costed, and the evaluation harness that makes every later decision checkable.

$16,000

2 weeks

160 hours at $100/hour

Pricing status: confirmed. PinkCloud quotes from these figures.

Prices exclude any applicable tax.

Deliverables

A written deliverable, named in the proposal before the work starts
An honest assessment of what is salvageable, with the repair path and the replace path both costed
An architecture review of the system or the design
Harness shape — a test suite for behaviour: task-level success criteria, regression cases drawn from real failures, and adversarial cases from the security practice
Harness shape — the suite running in CI, with results that move when the model, the prompt or the retrieval corpus changes
Harness shape — running code in your repositories, under your review process, with your team involved throughout

Exclusions

The practice's longer build shapes: the Retrieval pipeline (4–8 weeks), Agent orchestration (5–10 weeks) and Inference infrastructure (3–6 weeks), which are quoted as engagements at a weekly rate
Staff augmentation: fixed-scope work with an end date; if what you need is two engineers embedded for six months, that is a hire, and we will say so
Incident response and on-call cover: we are not an incident-response retainer and we do not carry a pager
A compliance certificate or an audit opinion: a larger firm with an audit practice will serve you better and cost less
Compliance evidence deliverables: these add 10 percent to the floor
On-site attendance: delivery is remote by default, and on-site work needs a scoping call
None of the above is a refusal: everything here has a route and a rate. Work outside this scope is quoted at our published hourly rate — as a change request during the engagement, or as a separate engagement afterwards — and is agreed in writing before it starts. The assessment itself is an agreed budget of 160 hours at that rate

Order

Questions we get asked

Do you build production systems or just advise?: We build. Engagements produce running code in your repositories, under your review process, with your team involved throughout — the goal is that they own it fluently when we leave. We are not interested in becoming a dependency you cannot remove.
Which models and frameworks do you work with?: We are deliberately model-agnostic and design for substitution, because the frontier moves faster than most procurement cycles. In practice that means an abstraction at the inference boundary, an evaluation set that makes swapping measurable, and no framework in the critical path that we would not be willing to fork.
Why does the security practice touch the build work?: Because retrofitting authorization onto a retrieval pipeline is far more expensive than designing it in, and because the failure modes we spend our time exploiting are cheapest to prevent at the architecture stage. Build engagements get security review as part of the work, not as an upsell.
Can you take over an existing system?: Yes, and it is a common starting point — usually a prototype that reached production faster than intended. The first phase is an honest assessment of what is salvageable, which occasionally concludes that a component should be replaced rather than repaired. We will say so plainly and cost both paths.
What size team do you field?: Small and senior. Boutique is a deliberate constraint, not a stage we are trying to grow out of — it is why the same people who scope an engagement are the ones who deliver it. It also means we turn work down when we are full, and will tell you that rather than staffing it thin.