Engineering notes

Articles

What we learned building software for companies. Written for the person deciding, not the person coding.

The client’s site got five times faster, and it cost nothing

Two switches left off in panels the client already paid for. Time to first byte fell from 1.3 s to 0.26 s, and the worst case from 14.7 s to under a second.

From 2,565 to 17,442 businesses on the map, without anyone typing one

Two public files from the Brazilian government solve a registry that would take years to build by hand. The work is not in the download, it is in deciding what to do when the address does not match.

One codebase, five cities: what changes when a system becomes a product

A marketplace built for one city became the base for five. The difference between a system and a product lies less in the code you write and more in the code you refuse to write twice.

Measuring the mood of a live chat is harder than it looks

Sentiment analysis works well on product reviews and badly on broadcast chat. ChatMood came out of that gap, and most of the work was not in the model.

I reorganised my project tool by client, not by repository

I had one tab per repository and could not answer the simplest question of the day: what do I still owe this client? The change took an afternoon.

Data protection and AI agents: what to settle before switching the bot on

Who answers for the data when an AI is doing the talking, what has to be in the vendor contract, and the practical checklist before a conversational agent starts serving customers.

AI customer service on WhatsApp: what it costs and when it pays for itself

What it costs to put an AI agent on your company's WhatsApp, the difference between a menu bot and a real conversational agent, and the arithmetic that shows when it pays back.

Process automation with n8n: what you can take off your team’s plate

How to pick the first process to automate, what n8n handles well, where it is the wrong tool, and why an unmonitored workflow is worse than a manual process.

From MVP to product: validating a SaaS without burning the budget

Most of the money lost in SaaS goes into building features nobody asked for. What to cut from the first release, what never to cut, and how to know whether validation worked.

ERP, CRM and calendar: how to stop typing the same thing three times

Re-typing data between systems is the most common invisible cost in mid-sized companies. How to integrate without becoming hostage to a vendor and without breaking on every update.

Legacy systems: full rewrite or refactor in place?

A full rewrite is the most expensive decision a company makes about software, and it is almost always made for emotional reasons. How to judge it coldly, and the alternative that usually works.

Multi-vendor marketplaces: payment splits, payouts and the mistakes that get expensive

What separates a marketplace that closes the month cleanly from one that becomes a manual reconciliation spreadsheet: splitting at the source, an explicit payout cycle and refunds thought through early.

Moving your system to the cloud without stopping the business

The steps of a cloud migration that does not turn into an all-night cutover: an honest inventory, containers before orchestration, data last, and a rollback plan written down first.

Enterprise blockchain: when Hyperledger Fabric earns its place and when it is just cost

A three-question test for whether your use case really needs a permissioned blockchain, or whether a database with an audit log does the same job for a tenth of the cost.

Monolith or microservices: how to decide without breaking the product

Microservices solve an organisational problem before they solve a technical one. How to tell whether your company has that problem, and what to do while it does not.

How much does custom software cost, and why do quotes vary so much?

Why three vendors quote three wildly different prices for the same system, what actually drives cost up, and how to ask for an estimate you can compare.

Next step

Need this solved at your company?

A twenty minute call is usually enough to tell whether it makes sense, and you talk straight to the engineer who builds it.