Skip to main content
Skip to content
How we work

Discovery, proof of value, production. Then again.

Every engagement is fixed and staged, priced in euros, under one agreement. The cycle below runs with every customer; the four chapters after it are what we do on every round.

Book a callSee the packages
01AI DiscoverySprintTwo weeks · fixed fee02Proof of valueWeeks · fixed scope03Adoption andintegrationStaged · one agreementTHEN AGAIN
CH.03 · How we work

Discovery, proof of value, production. Then again.

One cycle, every customer. Results in weeks. Each round starts from the numbers the last one produced.

Then round again, with what we measured.

DiscoveryProof of valueProductionMeasureevery customerevery round
  1. 01

    Discovery

    two weeks

    The problem in one sentence. Your data, seen. One prioritised backlog and a go or no-go.

  2. 02

    Proof of value

    weeks

    A working agent or skin on your data. Failure cases written down before anyone trusts it.

  3. 03

    Production

    installed

    On your systems, under your permissions, with a trail. We stay until it runs without us.

  4. 04

    Measure

    always

    What changed, in words and in the numbers you already track. The next round starts here.

CH.04 · Consulting

A decision with an architecture behind it.

Which layer, which runtime, which team goes first, what stays local. You leave with an architecture and a reason for each choice, then a go or no-go.

  • AI Discovery Sprinttwo weeks, fixed fee. One prioritised backlog. A go or no-go.
  • Solution architecturemodel, runtime, connectors. Where the data may go and where it may not.
  • Governance designwho reaches what, under which limits, with what record.
AI Discovery SprintTalk to an engineer
StrategyArchitectureGovernance
decideAI Discovery Sprint
CH.05 · Implementation

We build on your systems and stay until it runs.

One team covers machine learning, generative AI, agents and skins. The first deliverable is a proof of value on your data, with the failure cases written before anyone trusts it.

  • Proof of valuea working agent or skin on your data, in weeks. Failure cases first.
  • Machine learningchurn, forecasting, federated learning when the data cannot move.
  • Generative AIRAG, agentic workflows, fine-tuning, local LLMs.
Proof of valueTalk to an engineer
Machine learningGenerative AIAI SkinAgents
buildProof of value
CH.06 · Support and maintenance

Models change. APIs move. Someone answers.

Rules drift and vendors ship breaking changes. A named engineer absorbs them before your users see them, and Atalaia keeps the estate in view for the people who decide.

  • Model and API changesabsorbed and tested before they reach your users.
  • Atalaia on the estatecosts, permissions and drift, reported to the people who decide.
  • A named engineerone person who knows your systems, on the phone.
Adoption and integrationTalk to an engineer
OperationsAtalaia
runAdoption and integration
CH.07 · Training

The skin keeps growing after we leave.

Developers learn to code with agents against your house standards. Business teams learn to work inside the assistant. Leadership learns what to govern and what to leave alone.

  • Developerscoding with agents, against your house standards.
  • Business teamsworking inside the assistant, with the skills the skin ships.
  • Leadershippositioning and governance workshops for the people who decide.
Training pathwaysTalk to an engineer
DevelopersBusiness teamsLeadership
teachTraining pathways
CH.04 · Where it runs

Where it runs is a decision we make with you.

The same build runs in three places. Your compliance rules, your data and the tools your teams already have decide which.

Your assistant

Inside the one your teams already use

Claude, Codex or Copilot. No new application to open. No assistant yet? We choose one with you.

Local models

On your hardware, when compliance requires it

Open models, installed and tuned inside your perimeter. Nothing leaves the building.

Cloud, on our terms

Through OpenRouter, only models with no data retention

Pinned to the region your data must stay in. Europe by default.

Where your data may go is settled with you before the first line of code.

CH.08 · What we need from you

Four things, and we start.

Every ask on this list is something only you can bring. The boxes are drawn empty on purpose.

A planning session with a client

  1. A sponsor who can say go or no-go

    One person who decides, two hours a week during discovery.

  2. Read access to the systems in play

    The ERP, the CRM, the mailbox. Read only; nothing is written until production.

  3. One process you already suspect

    The job five people do five ways. We start there.

  4. The numbers you already track

    We measure against them, not against new ones invented for the report.

CH.09 · Packages

Three ways to start, each with a price and an end date.

Start

Bring us the problem nobody has cracked yet.

Small, senior, specialised. The right model, the right layer, the least machinery. A call with an engineer, no sales deck.

Book a call