Codin
Review a project Review

AI & Cloud Delivery Pod for consultancies, integrators and agencies

The AI & cloud lab that ships to production.

We build AI agents that hold up against real data, permissions and security controls. And we build the cloud platform that deploys them, observes them and keeps the cost in check. We join when there is a PoC in motion, a sold project, or a platform that needs sorting out.

Let’s talk See our work

A first 30-minute call, no commitment. We tell you what production still needs, whether you work with us or not.

100%
senior team,
end to end

No pyramid: the people who design the solution are the ones who build it.

Anonymized cases

What changes when the system actually reaches production.

Three anonymized deliveries. Each one shows the starting point, the window, what was built and the result—without turning one project’s number into a universal promise.

Legal SaaS - live product / 01

A new format stopped being a project.

Their product already extracted corporate documentation, but every new format meant retraining classifiers or writing rules. We rebuilt the pipelines as agents on language models: an unseen document takes the same path as any other, every field arrives with its source page, and anything inconclusive goes to a person.

Window: Phased delivery, format by format

100%
extracted fields carrying their source document and page
0
classifiers to retrain for a new format
Read the case
Financial SaaS - live product / 02

The engine that bills, rewritten while it ran.

A SaaS already billing on a calculation core with no tests and no typing, where every change was checked by hand. We rewrote it without stopping the product, demanding identical output against the previous engine, and left it on Azure with separate environments and rollback deployments.

Window: Phased rewrite, cutover at the end

3
separate environments: development, staging and production
0
differences allowed against the previous engine
Read the case
Tax SaaS - full platform / 03

The whole platform: code, cloud and identity.

A tax SaaS that had grown on a monolithic backend, a separate back office and standalone forms. We built the complete platform: domain services that talk through events, shared identity and roles, customer portal and back office, on Kubernetes in Azure with infrastructure as code and GitOps deployment.

Window: Built in phases, replacing the legacy system

6
backend services deployed independently
3
web applications on the same identity and roles
Read the case
Send us a project for review
The offer · AI & Cloud Delivery Pod

What we do and how we work.

You hire a delivery, not a collection of profiles: one senior pod combining applied AI and cloud platform engineering, embedded in your delivery and accountable through production on Azure.

  • Closed scope

    Scope and assumptions in writing before the first line of code. If something material changes, it is renegotiated under change control.

  • Biweekly milestones

    Every two weeks: a demo on the project's data and acceptance against the agreed criteria. Working software, not progress reports.

  • Exit after the first milestone

    If the first milestone misses what was agreed, you can walk away without carrying the rest of the engagement. Everything built already lives in your repositories.

  • Per-project price

    A fixed price against this scope. We put it in writing after the initial review, once we know what has to be built.

  • Bounded guarantee

    If something caused by our delivery fails during the agreed period, we review and fix it. Either ending — operate or hand over — leaves runbooks, the evaluation report and training.

The eight weeks

  1. Weeks 1–2

    Architecture and first deploy

    Review of what was sold or the existing PoC, the architecture of the agent and of the platform in writing, and a first deployment to the agreed environment.

  2. Weeks 3–4

    Build and integration

    We build the agent or the automation and integrate it against your systems: data, permissions, APIs. Evaluations run on every change from this stretch on.

  3. Weeks 5–6

    Hardening

    Guardrails, traceability for every response, cost and latency limits, and what the end client's security review will ask for.

  4. Weeks 7–8

    Production and handover

    Final deployment, runbooks and training. Then you choose: full handover, or continued operation within agreed hours with an improvement cycle.

The checklist of what production still needs

What you keep

  • The agent or automation in production, integrated with your systems
  • Infrastructure as code and pipelines with deploy and rollback, on Azure
  • Automated evaluations with a deployment threshold
  • Observability: traces, cost per task and latency
  • Runbooks, evaluation report and handover to your team

If the project doesn't fit in eight weeks, we say so on the first call and size the plan before signing.

Send us a project for review
Manifesto

Three conditions for calling a system production-ready.

A demo proves the idea can work. Production means being able to measure it, operate it and explain what happened when something fails.

  • 01

    Production is a standard, not a date.

    An agent is not in production just because it shipped. Evaluations, guardrails, traces and observability must make it measurable and operable. Every milestone closes against criteria written before we start.
  • 02

    The agent and the platform are one delivery and one team.

    The agent and the platform are designed as one delivery. You do not have to coordinate an AI vendor with a platform engineering vendor or translate decisions between teams.
  • 03

    Your brand, your client, your code.

    When the end client is yours, we work under your brand, with NDA, RACI and non-circumvention in writing. The code lives in your repositories, and the close includes runbooks and handover.
Who does it

You talk to the people who will write the code.

The same people on the first call design the agent, operate the platform and own the criteria for every milestone. There is no commercial layer translating between the people selling and the people building.

  • Rafael González, AI Engineer en Codin

    Rafael González

    AI Engineer

    Designs and builds the agents: retrieval over real data, evaluations on every change, guardrails and traceability. He is the one who answers when someone needs to explain why the system said what it said.

    LinkedIn
  • Martín González, Cloud Architect & DevOps Engineer en Codin

    Martín González

    Cloud Architect & DevOps Engineer

    Designs and operates the platform: Kubernetes, CI/CD pipelines with rollback, infrastructure as code, deployment automation, and observability for cost per task and latency. When the cloud bill or the deployment time has to come down, this is who answers.

    LinkedIn

  • No pyramid

    Technical ownership is not passed down a chain of profiles or rotated between accounts.

  • One single pod

    The same pod delivers the agent and delivers the platform. There are no two vendors to coordinate.

  • Decisions in one conversation

    No committee and no translation between whoever sells and whoever executes: what is agreed on the call is what gets built.

We bring in trusted collaborators when a project calls for it, and we say so before we start.

Who we work with

Who we work with.

Three entry points into the same difficult stretch: turning something that works at small scale into a system that can handle real data, users and controls.

01 - Consultancies

Consultancies, integrators and agencies that already sold the project

Where it stalls

You sold an agent or automation project and you don't have the pod to take it into production: it needs agents, LLMOps, Azure and Kubernetes at once, and those profiles don't line up on your bench.

How we join

We embed inside your delivery under your brand, with an agreed RACI, an NDA and non-circumvention in writing. The end client relationship stays yours; we sign the acceptance criteria for every milestone.

02 - Product

Product teams that already have users inside

Where it stalls

You have a product in production and an AI feature that shines in the demo and falls over on real data.

How we join

We come in to make it measurable before you open it to everyone, and to let you say why it answered what it answered.

03 - Internal systems

Companies whose own systems run the business

Where it stalls

You have an ERP, ticketing, a catalog or internal documentation the operation depends on.

How we join

We connect agents to those systems within your permissions, with a human in the loop where it matters and a clean exit: code in your repositories, runbooks and handover.

We fit when there is a concrete scope, someone able to make decisions and a date by which the system needs to be working. We do not sell hourly catalogs of profiles.

For consultancies, integrators and agencies

If the project belongs to your client, we come in under your brand.

NDA, RACI and non-circumvention in writing before the first line of code. The commercial relationship with your end client stays yours: we do not talk to them without you, and we do not turn your project into our case study.

How we work with consultancies
FAQ

What people usually ask before the first call.

If your question is not here, write to us and we will reply the next business day.

01 Can you work white-label inside one of our projects?
Yes, and we sign for it. NDA, non-circumvention and one simple rule: the commercial relationship with your end client is yours, and we sell them nothing on our own account, during the project or after it. We can appear under your brand in meetings, documentation and deliverables, or stay behind you and talk only to you. We agree beforehand who communicates what and how an incident is escalated. White-label work has its own terms, because we give up any public reference, and we put them on the table from the first conversation.
02 We already have a PoC built by another vendor. Do you start from scratch?
Almost never. We start by reviewing what exists: what it actually does, on what data, and what breaks when real users, permissions and volume arrive. That gives a short list of what we keep, what we rewrite and what is missing for production: evaluations, guardrails, traceability, observability, cost-per-task control and reproducible deployment. We work on someone else's PoC without putting whoever built it on trial. And if the approach will not hold, we say so before signing.
03 We are a large organization. Aren't you too small for this?
It depends on what you need. If you are after a program with dozens of people working in parallel, that is not us, and we will tell you on the first call. If you need a senior pod to own a specific piece and put it into production, the size works in your favor. You talk directly to the people who implement, and technical ownership is not passed down a chain of profiles. The contract covers the rest: closed scope, milestones with acceptance, an exit option at the first one and a continuity plan in writing.
04 How do you charge?
We usually price by project once the scope is understood, against milestones with acceptance criteria. In some cases we work through a monthly collaboration if continued technical capacity is needed on what we already delivered. We do not sell loose hour bundles or packaged services without reviewing the project first.
05 What happens with code and data?
The code lives in your repositories and remains yours from day one. We work with minimum access, agreed environments and confidentiality agreements. If we use AI tools, we define in advance which data can be used, where and within what limits, and we put it in writing so you can show it to your client or your security team.
06 Are you official partners of Microsoft, OpenAI or others?
We regularly work with Azure, OpenAI, GitHub, Kubernetes and Terraform, and we pick whichever tool each case calls for rather than the one a commercial agreement would oblige us to use. If your project or your client requires a specific official partner, tell us on the first call and we discuss it from the start.
07 Where are you based?
We are in Huelva, Spain, and mainly work remotely with clients across Spain and Europe. For kick-offs, architecture sessions or meetings with your end client, we can travel when it makes sense.
Next step

Shall we talk about your project?

In 30 minutes we review what you want to build or what you already have running, the data, the integrations and what it still needs to remain reliable once users arrive. Whether the project is yours or a client’s, the call is the same.

Free 30-minute review, no commitment. We sign an NDA beforehand if you need it. We reply the next business day.

or email us directly: hola@codin.es