Production AI engineering, from prototype to controlled regular use.

Radian IT helps teams take AI workflows from prototype to controlled regular use. We design the workflow boundary, build the production slice, define the evaluation fixtures and release controls, and hand over a runbook the buyer's team can operate from.

Available for one-to-two engagements at a time.

Inside production AI

Radian IT works inside production AI from the workflow boundary outward. The starting question is what the workflow is doing for the buyer, where it has to integrate, what the release owner needs to see, and what evidence makes a release defensible. Model and tool choices follow from those answers. That ordering is the difference: architecture, evaluation, and operating controls inside production AI, not model selection.

Production AI workflow build

The anchor engagement. One bounded workflow, six weeks, a single architect, and one deliverable pack. Fixed scope, fixed duration. From GBP 25,000.

When this fits

A useful AI workflow, prototype, or AI-assisted manual process works in demonstrations but cannot yet run as part of regular operations, because the workflow boundary, source-system access, tool authority, approval points, evaluation evidence, telemetry, release ownership, rollback, and runbook are undefined.

What you get

One workflow build pack: a deployed service in your environment, plus the workflow boundary map, evaluation fixture suite, run-evidence schema, release-gate rubric, deployment handover notes, and a runbook your team can operate from. Related proof: from AI prototype to controlled workflow.

Other engagements

Three further engagements sit alongside the workflow build. Each is fixed in scope and duration, with a single deliverable pack.

Agent memory and retrieval architecture review

Agent or RAG behaviour is inconsistent, hard to debug, over-reliant on prompt context, or unsafe to expand. You get an architecture review, retrieval and memory boundary design, a risk register, an evaluation outline, and an implementation plan. One bounded system, two weeks. Fixed scope, fixed duration. From GBP 8,000. Related proof: why agent memory needs architecture before autonomy.

OpenClaw and Hermes managed deployment

You want to operate an open agentic-stack runtime without owning the whole deployment path. You get a deployed and hardened runtime in your environment, with integration, tool policy, memory and channel setup where appropriate, handover notes, and a runbook. One named runtime into one environment, three weeks. Fixed scope, fixed duration. From GBP 18,000.

Agentic governance, observability, and evaluation audit

An agentic workflow is nearing regular use before tool authority, approval points, audit evidence, costs, or release gates are clear. You get a workflow and tool-authority map, a telemetry and audit-evidence gap analysis, a fixture outline, a risk register, release-gate recommendations, and a control roadmap. One bounded workflow, two weeks. Fixed scope, fixed duration. From GBP 10,000. Related proof: how to evaluate agentic workflows before rollout.

How we work

Direct first contact

Email what is happening, what needs to change, and any deadline or constraint. You get a written reply, not a forced booking funnel.

Small first step

Each engagement opens with a scoping checkpoint that bounds the work before the build phase starts.

Built where you can own it

Where possible, the work lives in your codebase, cloud, and normal way of operating.

Clear handover

The work ends with a deployed service, evaluation fixtures, release controls, notes, and a runbook your team can run or extend.

Where we work

Client references are kept anonymous and plural by policy. Sectoral patterns may be described; specific organisations are not named.

Government and public services

Public-service estates where approvals, evidence, and controlled rollout have to be designed in early.

Telecoms, media, and consumer platforms

Consumer-facing platforms where AI workflows touch high-volume, customer-visible operations.

Healthcare and regulated platforms

Regulated environments where data handling, audit evidence, and release control are not optional.

Scale-ups and software product teams

Product teams moving a useful AI prototype towards something they can run and own.

Start with the workflow.

Email [email protected] with what the workflow does now, what needs to change, and any deadline or constraint.