Get in touch

Fixed scope. Working software.

A working AI proof of concept in six weeks

One use case, your real data, your systems. We agree what success means before we write a line of code, and you end with working software and an honest evaluation, not a slide deck.

Book an intro callSee the scope

Trusted by 40+ businesses

The package

Six weeks on one use case, with a clear answer at the end.

The exact scope depends on integration depth and data complexity. We fix it, together with the quote, on a scoping call before the project starts, and neither moves afterwards.

AI Proof of Concept

6 weeks
  • A working PoC running on your real data and connected to your systems
  • Success criteria agreed upfront and an evaluation report against them
  • Production architecture proposal with running-cost projections
  • Source code, documentation, and full IP ownership on your side
  • A go or no-go recommendation you can take to your board
Scope your PoC

Success criteria we agree before we start

Scope your PoC

Accuracy on your data

A measurable quality target, defined with your domain experts and evaluated on real cases, not on a curated demo set.

Cost per request

A budget for what each interaction may cost in production, so the PoC proves economics, not just capability.

Latency your users accept

Response-time targets that match the workflow the AI sits in, measured under realistic load.

Integration, proven

The PoC talks to your actual systems, so the hardest question about production readiness is answered early.

The six weeks, week by week

A fixed cadence with your team involved throughout, not a black box that opens at the end.

  1. 01

    Scope and criteria

    Week one. We pick the narrowest slice that proves the value, get data access, and write down the success criteria together with you.

  2. 02

    First working version

    Weeks two and three. The solution runs through the whole process on real data: still unpolished, but working. You see it as it develops and can request changes before we go further.

  3. 03

    Evaluate and iterate

    Weeks four and five. We measure against the criteria, fix what falls short, and involve your domain experts in reviewing outputs.

  4. 04

    Harden and hand over

    Week six. Documentation, evaluation report, production architecture with cost projections, and a readout with a clear go or no-go recommendation.

What is in scope, and what is not

A PoC earns trust by being honest about its edges.

In: one use case, end to end

A single workflow proven from input to output, connected to real systems and evaluated on real data.

In: your infrastructure when needed

EU-region cloud by default. On your own infrastructure or with self-hosted models when your compliance requires it.

In: evaluation you can trust

A test set built with your experts and results reported against the agreed criteria, including the failures.

Out: production rollout

The PoC proves feasibility and economics. Scaling, SLAs, and organisation-wide rollout are the next project, and the architecture proposal prices it.

Out: several use cases at once

One question answered well beats three answered halfway. If you have several candidates, the AI Readiness Audit ranks them first.

Out: unrealistic promises

If the data cannot support the use case, you learn that in week two, not in month six. That is what a PoC is for.

What you know when the pilot ends

Scope your PoC

Whether to build it

A decision based on test results on your own data and in your own systems, not on a vendor demo.

What it will cost to run

A running-cost projection based on real usage: infrastructure and model fees, before you commit to scaling.

What the path to production looks like

A production architecture proposal: what needs to be built, in what order, and at what effort.

What is yours

The code, prompts, test sets, and documentation belong to you. Continue with us or with your own team.

From PoC to production

Systems that started as a pilot to test an idea and grew into products.

Genie AI: Shopping Advisor
AI Assistant PlatformConsumer & Retail Tech

Genie AI: Shopping Advisor

An AI assistant for a UK shopping app: it answers questions about real spending, draws charts in the chat, and never invents a number.

View Case Study
Genie Vision: AI Campaign Targeting
AI Product DesignRetail Media & Analytics

Genie Vision: AI Campaign Targeting

A campaign product for retail chains: marketers build precisely targeted offers through a conversation with AI.

View Case Study

Before you ask

You get a production architecture proposal with cost estimates as part of the package. If the PoC meets its criteria and you want to continue, that proposal becomes the plan. If you continue with another team, everything they need is in the handover.
You do, fully and from day one. Source code, prompts, evaluation sets, documentation. There is no lock-in by design.
Two to three senior engineers, including one who has shipped LLM systems to production. No juniors learning on your budget, no handoffs between departments.
You get the evaluation report explaining exactly why, and what would have to change for the use case to work. A clear no with evidence is a cheap outcome compared to discovering it after a year of building.
Whatever fits the case: OpenAI, Google, Anthropic, or self-hosted open models when data cannot leave your infrastructure. We build provider-agnostic, so switching later is a configuration change, not a rewrite.

Let's Talk About Your Project

Get In Touch
Maciej Roman|CEO & Co-founder

Not sure where to start? Begin with the AI Readiness Audit