Samo Fortuna · design engineer

Convincing people and machines to do what’s actually required.

I take a vague need, write it up as a spec someone can build from, and check that what comes back, from people or from models, is what was asked for.

The shift

Models write the code now, and someone has to judge it.

A good spec is close to executable these days. Generated code comes fast and is often wrong in ways that only show up later. So most of my time goes into deciding what should be built, finding where it will break, and fixing it when it does.

Background

Thirty years of stack, so I don’t take the model’s word for it.

I started with PHP 4 on Solaris and took up serverless on Azure early. I’ve worked at every layer of the stack, so when a model suggests something, I’ve usually built that or the alternative and know where it breaks. My MSc in Management of Software Projects from the Open University is why the business case comes first.

1. Business

   Requirements, costs, and whether the thing should exist at all. Requirements captured with Volere. Years of working with lawyers and business owners. Documents in LaTeX.

2. Team

   Leading and managing developers. I saw the value in Netlify CMS, now Decap CMS, and put together the team that maintains it today.

3. Delivery

   CI/CD on every project. There’s no excuse to run without it. GitHub Actions by preference.

4. Security

   Treated as a whole. The CIA triad: confidentiality (only the right people read the data), integrity (nobody changes it unnoticed), availability (it’s there when needed). All three are requirements, and Volere has a place for each. A firewall covers a small part of one. With models holding close to root access to everything, you can only manage the risks you can see, so look wide.

Stack

From the cloud down to the circuit board.

1. Cloud

   Azure since the early serverless days: Functions, Durable Functions, Service Bus, Event Grid, Cosmos DB, Container Apps, App Service, Static Web Apps.

2. Application

   PHP since 4.x, C# from .NET Framework to today’s .NET, Node.js, TypeScript, Python. Vue 2 and 3, and jQuery before that. Plenty that never went near a cloud, too: ERP and utility apps, console and WinForms.

3. Data

   Mostly SQL Server, in depth: T-SQL, user-defined types, efficient indexing. Still my first choice; I don’t know a more complete database, and it just works. A lot of MySQL, some PostgreSQL, Redis, search engines.

4. System

   Unix before Linux, starting on Solaris. Linux administration ever since. Notes on what a system administrator should be doing.

5. Hardware

   STM32 and ESP32 firmware on ESP-IDF, in C and C++. SystemVerilog on Zynq. Test setups driven over SCPI. PCBs from TangoPCB to KiCad. Over three years on the Noordung Boombox e-bike: regulation, a team from several disciplines, and a production run too small for the hardware world to take seriously. Notes on getting a hardware product through CE.

I don’t use Kubernetes, on purpose. For most small and mid-sized companies it costs more than it gives back.

Contact

Have something vague that needs to become real?

Tell me what you’re trying to do, and we’ll work out what it really needs.

Email works too. It’s easy to guess from my name and this domain.

Nabla, Samo Fortuna s.p.