I work on integrations between machines in the field and the systems companies run on. I take projects as a consultant: I start from the process I find, size the solution to it, and stay the person to call when something stops.
I read the data collected on board the machine and write it into the ERP. The job runs every day without an operator. Every record keeps its source and the result of the write, and the link between machine and ERP stays documented.
python, rest, sftp, t-sql, sql server
See the case →I build the application and take care of the release, with staging environments kept separate from production. Database migrations go in two phases: the reversible part first, the final part only after verification on real data.
typescript, next.js, fastify, postgresql, prisma, docker, ci/cd
See the case →I take a process done by hand today, break it into verifiable steps and hand it to a program. The cost of every run stays under control.
This line of work is at its first pilot project.
typescript, anthropic sdk, node, jsonl
See the internal R&D →Data from the sixty vehicles stayed inside the vendor portal. An operator copied it into the ERP by hand. Nobody else knew the procedure, and no history was kept.
When the workload was high, some days were skipped. Management control saw spikes in fuel consumption and mileage weeks late, too late to dispute anomalies or wrong supplier charges.
In production, it runs on the server and the scheduler the client already had, with no extra paid services. Every run leaves a diagnostic log, an error sends an email notification.
| By hand | Automated | |
|---|---|---|
| Frequency | days skipped when the workload was high | every day |
| Consumption anomalies | seen weeks late | data updated every day |
| Procedure | known to one person | documented |
| History | none | in SQL Server, with source and result of every record |
| When something stops | noticed weeks later | email alert and diagnostic log |
python 3.12, sql server 2019, t-sql, rest, sftp, fernet, pytest
An application used every day to record attendance needs frequent updates, and every update touches data the client companies have already recorded.
A wrong migration is rolled back in the first phase, before it becomes final, and the release does not depend on who runs it.
typescript, postgresql, docker, ci/cd, git
On cogeneration plants, sampling boards and PLCs collect time series of thermal and electrical telemetry. This data cannot be recreated: if the database is corrupted or the storage fills up, that day's performance data is lost for good and the reporting required for state incentives is no longer valid. The installations run on hardware at the client site, on VMware and on remote servers, and a fault has to be isolated remotely.
Fault diagnosis got faster: the tools point to where the work is needed before anyone opens a session. Client data stays recoverable even after a database fault.
linux, postgresql, python, bash, vmware, digitalocean
An IT service provider runs first-level support on behalf of several companies, all on the same portal. Each company must see only its own tickets, and the provider must know which ones are about to miss the agreed times.
Separation between companies lives in the database, so it holds for every query the portal runs. Tickets close to their deadline surface before the time is up.
java, jdbc, postgresql, row level security
Internal project, developed outside any client engagement. No client runs it in production.
The part I work on is cost control. A shared token ledger measures every run and enforces a cap of $2.00: past that threshold the supervisor stops the workers, so a loop cannot turn into an unexpected bill. In internal tests, a full cycle of analysing and breaking down an average process costs about $0.07.
Agent orchestrator developed in TypeScript directly on the Anthropic SDK, with no intermediate framework.
A supervisor breaks the work down and assigns tasks to workers. State persists to JSONL files and filesystem tools operate inside a sandbox.
Development continues in order to test automation on real processes. Once a project of this kind runs in production at a client, I will document it under work.
typescript, anthropic sdk, node, jsonl
Which machines or systems are involved, and what has to happen. I will tell you whether it is work I can take on and how I would set it up.
I work across Lombardy, Veneto and Emilia, and remotely. Based in Mantua.