Systems in operation

These are my employer’s systems. I tell the problem each one solves and what my role was; internal architecture and operational detail stay out.

Industrial monitoring

I structured the system: architecture, backend and frontend.

An OEE (overall equipment effectiveness) platform that reads machine telemetry and returns indicators in real time to whoever operates. The terminal the operation uses during the shift keeps working without a network, and what counts as downtime is a rule configurable by whoever operates, not code.

.NET · MQTT · TimescaleDB · SignalR · React with offline PWA

Payments and anti-fraud

I am technically responsible for the platform. I have worked on it since 2023, and since November 2025 I answer for architecture, contract, technical decisions and review.

The central payments platform. It manages the available methods, orchestrates the path of each transaction, including when an order is paid with more than one method, and concentrates a good part of the anti-fraud decision. It is the most critical system under my responsibility.

.NET · Oracle · Entity Framework Core · Dapper

AI engineering tooling

I built it, and it stopped being mine: today other people in engineering commit to it.

Engineering tooling around the AI agent. The house tools became commands the agent operates, the standards became instructions it loads by itself, and day-to-day friction became the input of a cycle that generates new tools. The design and the decisions behind it are in AI engineering.

Tender triage with an LLM

I built most of the system. It is where I put AI inside the product, and not in the process of making software.

The national public procurement portal publishes thousands of tender notices a day, and someone needs to decide quickly which ones are worth bidding on. A language model reads the notices and gives each one a relevance score with the reason for that score; the system shows the ranked queue and records the verdict of whoever decides. The model ranks, the person decides.

.NET · Oracle · MediatR

Buyback

I rebuilt a good part of a system that already existed.

The customer hands in a used device as part of the payment for a new one. To close the purchase they need a value now, but the real value depends on the condition of the device, which is only known days later, at inspection. What I rebuilt is the reconciliation between the two appraisals: an automatic outcome when they match, and a path of its own for each kind of divergence.

.NET · Oracle · Entity Framework Core

Commissions

I rebuilt the invoice extractor; in the rest, I contribute.

Commission calculation for several sales channels that pay in ways incompatible with each other, and that needs to be auditable: someone will ask, months later, why that amount came out the way it did. The extractor is the single point through which an invoiced sale enters the commission systems.

Customer service

I built the panel; the middleware behind it, I started and did not do alone.

The agent needs to know what happened with the customer’s order, and each answer lives in a different place, with its own portal and login. The panel runs inside the customer service platform itself, on the side of the ticket, and shows the status of the order without the agent leaving it.

React · Vite · .NET

Manufacturer reporting

I built the ETL.

A manufacturer requires periodic reporting by contract, and what leaves here has to arrive in its vocabulary. The ETL processes several kinds of submission per run, and a kind that fails does not take down the others: operations knows exactly what to reprocess.

.NET · MediatR · Oracle