Decision Blueprint · The first step

Decide what to build, before you build.

One decision, reviewed in your environment. You leave with a written plan: which kinds of AI, what the data allows, where the approval limit sits, what hardware it needs, and how it will be tested.

Request a Decision BlueprintTalk to the engineer

FIG. 1 · Decision BlueprintWritten planIllustrative

SME credit referrals: what do we build first?

Which kinds of AI
A model and an agent; the small language model waits
What the data allows
Past decisions usable; reason codes partly missing
Where the approval limit sits
The agent prepares; a credit manager signs every case
What hardware it needs
Your servers; existing capacity is enough
How it will be tested
The credit team grades past files before any build
Outcome
  • Build
  • Adjust
  • Pause
  • Stop

What is in the plan

Five answers about one decision, in writing.

FIG. 2 · Decision Blueprint · SME credit referralsIllustrative

SME credit referrals: what do we build first?

Which kinds of AI

  • A model and an agent; the small language model waits
  • Why: the statements arrive structured; there is no reading task yet

A machine-learning model, a small language model, an agent, any one on its own, or none, and why.

What the data allows

  • Past decisions usable; reason codes partly missing
  • Fix first: the reason codes on referred files

What the data can support, what is missing, and what must be fixed first.

Where the approval limit sits

  • The agent prepares; a credit manager signs every case
  • Inside the written limit: nothing changes

What the system may do inside your written policy, and where a case goes to a named person.

What hardware it needs

  • Your servers; existing capacity is enough
  • No outside model service

Your servers or your cloud account, the systems it connects to, and the hardware sized for the task.

How it will be tested

  • The credit team grades past files before any build
  • Go-live only on the tasks that pass

The test your expert writes, agreed before any build starts.

Outcome
  • Build
  • Adjust
  • Pause
  • Stop

How it ends

Every Blueprint ends in one of four outcomes, stop included.

Stopping is an honest result. If a rule or a ready-made product does the job, you have saved the build.

  • Build

    The problem is real, the data is ready, and the gain can be measured. A validation build comes next.

  • Adjust

    The right problem, with another kind of AI or a narrower scope.

  • Pause

    The data or the process must be fixed first, and the plan says what.

  • Stop

    A rule or a ready-made product is enough, and we say so in writing.

The plan is yours, whoever builds it.

How it runs

Four phases, inside your environment.

It typically takes 2–4 weeks. What makes it longer: more than one decision in scope, or data access that has to be arranged before the work can start.

  1. Problem and access

    We agree the one decision, who owns it, and how we reach your systems through your approved route.

  2. Data and process

    We read the data and sit with the people who work the desk today, to see what the data can support.

  3. Design and test plan

    We choose the smallest setup that does the job, where the approval limit sits, the hardware, and the test your expert will write.

  4. Written plan and readout

    You receive the plan and its outcome, walked through with the people who decide.

What we need from you

A few hours a week from the people who know the desk.

  1. One decision, and the person who owns it.
  2. Read access to the data and documents in scope, through your approved route.
  3. An expert who can judge the outputs and will write the test.
  4. An IT or security contact for access and hardware.

After the Blueprint

The Blueprint first; if the plan says build, two steps follow.

Each is a separate decision. Nothing in the Blueprint commits you to the next step, or to eqlx.

  1. Decision Blueprint

    Typically 2–4 weeks

    A written plan for one decision, and its outcome.

  2. Build · Adjust · Pause · StopYou decide

  3. Validation build

    Typically 4–8 weeks

    The key assumption tested on your own cases, against your expert's test.

  4. Go live?You decide

  5. Production, integration, handover

    Typically 3–6 months

    We build and integrate, or integrate what you already run: core banking, ERP and MES, CRM and the contact center, your decision engine, your document stores, your servers or cloud.

eqlx accessNamed accounts your team issues, time-limited, logged; no standing remote access.At handover: accounts closed, keys rotated.

What it is not

The Decision Blueprint is a technical review. It does not replace independent validation, certification, a cybersecurity audit, or legal and compliance advice.

A fixed fee, agreed before it starts.

Known questions

What buyers ask before they start.

How does the first conversation go?
You describe one decision at a high level. The engineer who would lead the work reads every message and replies with questions or a view on fit. No confidential data is needed at this stage, and the first conversation carries no charge. You can write in English or Turkish.
Can we start from a machine-learning model we already have?
Yes. The model is re-evaluated against what it is for, then recalibrated, monitored, and documented as the evidence requires; it is not rebuilt by default. A small language model or an agent is added only where the decision needs one.
Do we need a large GPU investment to run a small language model?
A small model trained on your task often runs on a single server in your own data center. We size the hardware in the Decision Blueprint, measured on your own cases, before you buy anything.
Who owns the work?
Customer data remains customer data. Customer-specific deliverables belong to the customer under the engagement terms. Pre-existing and reusable eqlx technology remains eqlx IP. Third-party and base-model licenses continue to apply.
Where is eqlx based, and where does the system run?
eqlx is based in İstanbul. Where and how we work together is agreed when the scope is set. Regulated work runs on your own servers, cut off from outside networks where required, and keeps banking data in Türkiye where the rules require it; other work runs in your cloud account or with a local provider you choose.

Equalize the gap.

Bring one decision. Leave knowing the first step.

A short description of the decision, who owns it, and where it runs is enough to plan the Blueprint.

Request a Decision BlueprintTalk to the engineer

https://eqlx.ai/decision-blueprint · Printed copy; the website is the current version.