Questions we get often.

An applied AI agency. We build production AI agents and workflows, deploy voice assistants on your phone lines, fine-tune models on your data, and run the evaluations that show whether a system is safe to ship. We are engineers, not a reseller of someone else's product, and we work on a small number of engagements at a time.

Operations, product, engineering and risk teams in fintech and healthcare: lenders, payment companies, wealth platforms, insurtechs, clinics, provider groups and digital health products. The common thread is a workflow where a wrong answer has a cost and someone has to explain it afterwards.

We are based in Belgrade, Serbia, and we work with teams across the UK, the EU and North America. The time zones overlap well with all three, and most of our work happens remotely inside your environment.

Fintech and healthcare. Both are regulated, both involve decisions with real consequences for real people, and both require evidence before launch rather than a convincing demo. That is the kind of work our process is built for. If your sector has the same constraints, we are happy to talk.

We publish case studies only with written client approval, and we do not show logos or testimonials without it. If you want references, ask on the scoping call and we will tell you honestly what we can share.

Usually a discovery sprint on one workflow, or an evaluation audit of an AI feature you already have. Both are fixed-price, both run in one to four weeks, and both end with a written readout you can act on whether or not we work together afterwards.

A discovery sprint is one to two weeks. An evaluation audit is two to four. Pilots typically run four to eight weeks, and full builds six to twelve depending on integrations and your review cycles. The timeline is written into the scope before work starts.

In your cloud account or on your infrastructure by default. We can host for small pilots if you prefer, but production systems in regulated environments belong with you. Where a third-party model API is used, we agree the provider, the region and the retention terms with you first.

You do. Integration code, infrastructure definitions, evaluation harnesses, datasets and fine-tuned weights (subject to the base model's licence) are handed over at the end of the engagement. We do not keep copies of your data or your models afterwards.

Every build ends with a runbook, monitoring, a regression suite and working sessions with your engineers so they can run, extend and re-evaluate the system without us. If you would rather we stayed involved, a monthly retainer is available on thirty days' notice.

When the task needs consistent format or behaviour across thousands of calls, when the domain language is far from general text, or when you need a smaller model for cost, latency or data-residency reasons. If prompting and retrieval would do the job, we say so in the discovery sprint and you save the training budget.

A sample of outputs is scored by your domain experts. We measure agreement between the automated judge and the humans, tune the judge until agreement is acceptable, and report that agreement rate alongside every automated score so you know how far to trust it.

On every change to the prompt, the model, the retrieval data or the guardrails, which is why the suite goes into your CI pipeline. A full red-team pass is typically repeated before major releases or when the threat model changes.

It produces the kind of evidence those processes ask for: documented test design, results, known limitations and mitigations. It does not make you compliant on its own; your assessors decide that. We write the report so they can use it directly.

Less than most people expect for behaviour and format, often hundreds to low thousands of high-quality examples. Teaching a model new knowledge needs more, and is often better solved with retrieval. The baseline measurement tells us how far the current model is from the target, and that drives the data plan.

Yes, on every engagement. It sets out what we process, on whose behalf, for what purpose and for how long, along with the roles and responsibilities on each side. Our standard DPA is published on this site and we are happy to work from yours instead.

Nowhere. Systems run in your cloud account or on your infrastructure, training runs on your GPUs or on rented capacity under your account, and we do not keep copies of your data or your models after hand-over. Where a third-party model API is involved, we agree the provider, region and retention terms with you first.

Our processes are designed to support your UK GDPR and EU GDPR obligations: a DPA on every engagement, processing limited to the agreed purpose, data kept in your environment, and special-category data handled under the stricter rules. Full compliance also depends on your own lawful basis and configuration, which is why the DPA sets the responsibilities out explicitly.

Standard deployments are not represented as HIPAA compliant. For US healthcare use cases we discuss whether a compliant path, with the required vendors, Business Associate Agreements and controls, is feasible for your situation before any work starts.

We do not currently hold SOC 2 or ISO 27001 certification, and we will not claim certifications we do not have. What we commit to in writing is how we work: a DPA on every engagement, your data in your environment, cardholder data kept out of model inputs and training sets, and evaluation evidence written for your assessors to use. If a certification is a hard requirement for your procurement, tell us on the scoping call and we will say plainly whether we can meet it.

Fixed prices for fixed scopes. Discovery sprints, evaluation audits, pilots and builds are each quoted as a single fixed figure, written into the proposal after the scoping call and before any work starts. Ongoing work runs on a monthly retainer with thirty days' notice.

An agreed amount of monthly engineering capacity for continuous evaluation, model and prompt updates with regression checks, a priority support channel and a quarterly review with your stakeholders. It runs month to month on thirty days' notice.

For builds: the number of integrations, the sensitivity of the data and the depth of your review cycles. For training: how much data needs collecting and labelling, and whether your experts are available to review it. For evaluation: the number of distinct behaviours to test, how much human labelling is needed to calibrate judges, and whether red-teaming is in scope.

Fixed-price engagements run to the end of their scope, and you keep everything produced up to that point. Retainers can be cancelled on thirty days' notice with no penalty. If the work is not earning its keep, we should not be in business with you.

No. Unpaid pilots push the cost of scoping onto whoever says yes first, and they rarely get the attention a real engagement does. The discovery sprint is the cheapest way to find out whether a project is worth doing, and you keep the readout either way.