Desemax · Backend and AI engineer · MilanDesemax · Senior backend and AI-integration engineer · Milan
Backends that hold up in production, including the ones AI wrote. Evidence attached.
Eleven years of PHP, Laravel and Symfony. AI designed into products, AI-built software rescued, legacy rebuilt against the live system, performance fixed with numbers. Independent since 2015.Eleven years of PHP, Laravel and Symfony. I design AI into products so it holds in production, rescue software built fast with AI tools, rebuild legacy against the live system, and fix performance with numbers rather than opinions. Independent since 2015, working across Europe.
What I do
Two kinds of work. The engineering is what I have done for eleven years. The AI work is newer, and built on it.
Engineering
Backend and performance
PHP, Laravel and Symfony systems that hold up under load. Slow pages fixed by measuring first.
Legacy rebuild
A system nobody wants to touch, rebuilt and checked against the live one, screen by screen.
AI
AI system architecture
AI designed into a product so it holds in production: retrieval, tools, guardrails, cost. The voice AI platform in case study 01 is the example.
Rescue of AI-built software
An app built fast with AI tools that now breaks. I audit it, write down what holds and what doesn't, and fix it.
Fractional Tech Lead
A part-time senior technical lead for a startup team: architecture decisions, reviews, and the code when it matters.
Case studies
Clients withheld. Every figure measured, or marked as not.
A voice AI platform in production. One engineer.
Phone, WhatsApp and web chat, live in sixteen production deployments.An AI assistant answering phone calls, WhatsApp and web chat, live in sixteen production deployments. The architecture, and the four decisions that mattered.
928 queries per page, down to 36.
An energy retailer's back office: page loads from 22.3 seconds to 2.7.Profiling and fixing the back office of an energy retailer: page loads from 22.3 seconds to 2.7.
Built against a mock. Proven against the national system.
A live check against SII, finished four months before access arrived.A live eligibility check against SII, Italy's national energy-market system, finished and tested four months before access arrived.
Rebuild against the live system, not the source.
Side by side, screen by screen, every divergence numbered.How I verify a legacy rebuild before anyone relies on it: side by side, screen by screen, every divergence numbered.
How I work
Three rules that show up in every engagement. They are also why the case studies read the way they do.
- 01
Real numbers, or none.
Every figure is measured and says where. If nothing was measured, the document says so.Every figure I publish is measured in production or in a test environment and says which. If nothing was measured, the document says so.
- 02
Verify against the live system.
Source says what was intended. The running system says what is true.Source code tells you what was intended. The running system tells you what is true. Rebuilds, integrations and migrations are checked against the second.
- 03
One engineer, accountable.
You deal with the person who writes the code.You deal with the person who writes the code. Architecture, delivery, the on-call call at the wrong hour: the same name on all of it.
Four ways to engage
You can change shape halfway: from a project to a retainer, or from a first module to a second.
A fixed-price audit
For software built fast with AI tools, or any system nobody fully understands any more. Security, data, structure and running cost, written down: what to keep, what to fix, what to rebuild. Then you decide.
Fixed price, agreed up frontA project
Defined perimeter, fixed deliveries. For replacing something that runs on spreadsheets, or on a system nobody will touch any more.
Quote after the first callA first module in 90 days
One process, into production, used by the people who own it. Enough to know whether working together works, before committing to the rest.
Budget agreed up frontA retainer
Maintenance, small evolutions and someone who answers, for a system that already runs. The same person next year.
Monthly, renewed yearly
An AI assistant for pharmacies, on phone, WhatsApp and web chat.
Answers as each pharmacy, from its own catalogue, services and hours. Live in 16 pharmacies across 8 regions.Answers as each pharmacy, from its own catalogue, services and hours. Built and run as a product; the engineering is in case study 01.
- Channels
- Phone · WhatsApp · web chat
- Live
- 16 pharmacies, 8 Italian regions
- Answers
- Real-time model, no orchestration layer
Contact
The first step is an email.
Tell me what runs today and what it should do next. I reply within one working day.Tell me what runs today and what it should do next. I reply within one working day, with questions, not a quote.
richard@desemax.it