Cases

Anonymized cases with what was measured. And with what was not.

Due to contractual confidentiality, no case is published with a client name, identifiable sector, or commercial value. Where no measurement is available, the case is marked as Illustrative, meaning the problem is real and the approach was applied, but the numbers cannot yet be attributed.

Four cases

From the problem to the declared state, with no invented number.

Two of them are illustrative, one is in production and verifiable by the visitor themselves, and one is in definition. The state comes before the title in every block: it is the first thing you read, and it is what keeps a technical design from being read as a delivery.

Illustrative

Industry · multi-country integration

An integration platform replacing dozens of point-to-point connections

Description
Operations across three countries with local systems talking to each other through direct connections. Each new partner was a new project. We designed and implemented a centralized integration layer with versioned API contracts, async messaging, and end-to-end tracing. Onboarding a new partner went from being a project to being a configuration on an interface that already existed and was already monitored.
Before and after: point-to-point connections become an integration platform Before, every new partner was a new project over direct connections. After, a versioned API contract, async messaging and end-to-end tracing make onboarding a configuration on an interface that already exists and is already monitored. BEFORE Direct linkspoint to point New partnerwas a new project AFTER API contractversioned Queue + tracingonboarding is config
Illustrative: the declared before and after, with no measured number.
Illustrative

Retail · data readiness

Data readiness before spending on AI

Description
Board pressure to launch an AI use case on a tight deadline, over a data environment with no documented lineage. Our readiness diagnostic (source inventory, lineage, quality with domain ownership, sensitivity classification and access control) showed that only one domain had trustworthy data. The pilot was scoped to start with that domain, and two blockers that would have killed the demo became visible before the model was purchased.
Before and after: data readiness before spending on AI Before, a data environment with no documented lineage and a tight deadline for an AI pilot. After, the pilot starts with one domain that has trustworthy data, and the blockers become visible before the model is purchased. BEFORE No lineagenot documented Tight deadlineAI pilot AFTER One domaintrusted data Pilot slicedtwo blockers visible
Illustrative: the before and after of the readiness diagnostic, with no measured number.
In production

ArquitechCorp infrastructure · institutional site

Brazilian cloud as a verifiable argument, not a promise

Description
Much of the market's offering for data residency in Brazil exists only in contracts. This site is served from Object Storage at Magalu Cloud (br-ne1) with Cloudflare in front. No third-party resources are loaded. The claim is verifiable at any moment: the response originates from Brazilian territory.
Before and after: data residency the visitor can verify Before, data residency exists only in the contract and in commercial material. After, this site is served from Object Storage at Magalu Cloud, region br-ne1, and the origin of the response can be verified by the visitor. BEFORE Residencycontract only A claimwith no public proof AFTER br-ne1Magalu Cloud Origin verifiableby the visitor
In production: on this site the origin of the response is verifiable by the visitor.
In definition

Brazilian enterprise · search risk and brand exposure

Brand exposure treated as a business risk, with register and owner

Description
A significant portion of what is said about a large company today is produced by search engines, aggregators, and language models, and is governed by no one inside the company. We built a proprietary framework that reads search exposure as business and compliance risk, with a risk register, escalation matrix, RACI model, and executive scoreboard. The first measured case will come from our own operation.
Before and after: brand exposure with a register and an owner Before, risk with no owner and no record. After, the framework reads search exposure as business and compliance risk, with a risk register, an owner and a deadline, and the first measured case will come from our own operation. BEFORE Risk, no ownerno register Exposureoutside governance AFTER Risk registeredowner and mandate Deadline, decisionexecutive board
In definition: the register and the owner are designed, and the first measured case will come from our own operation.

From the vault

Two real cases from the vault that support the newsletters

The two below are the ones with published material: the problem is real, the approach was applied, and the account is written down, with the questions that stalled the project and the decision that unblocked it, not just the final result.

Newsletter 04

Oracle to Databricks · data governance

I Took Part in a Project That Archived a Decade of Payroll Data

Description
A project archiving ten years of HR and payroll data from an Oracle system into Databricks. The questions that stalled the project for a week were not about the pipeline, they were about governance: where to censor sensitive data, who can see the unmasked version, how to prove control afterward. The lesson documented in Newsletter 04 is that readiness for AI does not start with a data catalog; it starts with who can see what, and whether that survives an audit.
Read on LinkedIn
Vault

Legacy VM · Azure · credentialless bypass

The Firewall Was Never the Problem

Description
A Linux VM from 2010 in production needed to send logs to Azure. Network worked, but SFTP would not connect: the VM's SSH client spoke a 2010 cipher dialect, and the Azure endpoint only accepts modern algorithms. The solution: a bridge VM that the team controls between the legacy machine and the cloud. The legacy sends via the old protocol; the bridge uses managed identity with Azure. No cloud credential ever touches the 2010 machine.
Read on LinkedIn

Neither of the two is an identified client case: what is published is the technical problem and the decision, without the name of whoever commissioned it. It is the same criterion applied to the four cases above, and it is why these two serve as the source material for the newsletter copy.

Want to see your case here?

Thirty minutes to describe the context. If it is an architecture problem, the next case could be yours, anonymized, like all the others.

No form in this version: a form here would mean processing personal data with no declared legal basis. Direct channel, response within one business day.