I work with businesses of every size, from the shop round the corner to a group of companies. If you are not sure which service you need, write to me: the first answer is free.
Automation and artificial intelligence
If somebody in the company spends their mornings reading emails and copying data from one place to another, most of that work can be automated. Not in theory: I have a platform in production that reads a tour operator's emails and works out the orders on its own, with 90.6% of fields correct on the first pass across 379 emails processed.
An automation that is not measured is not an automation, it is a hope. That is why the first step is not choosing a model: it is counting the hours that job takes away today.
Half a day watching who does what, with which documents and in how long. Not how it ought to be: how it is.
Hours per month, per person. That is the number that says whether automating is worth it and how much it makes sense to spend. Sometimes the answer is that it is not worth it, and I tell you.
Not everything at once: the part where most time is lost and most mistakes happen. Usually it is the copying from a document into a program.
I take your emails and your PDFs, not invented examples. It is the only way to know the accuracy before the system goes into production.
The system proposes and flags, it does not decide. Somebody approves, and every approval becomes an example that improves the next pass.
How many fields right, how many to correct, across how many documents. On Manifest that is 8,655 blocks extracted and 221 orders generated: those are numbers, not impressions.
One piece at a time, with the figures to hand. Automations that get extended everywhere at once are the ones nobody ends up using.
The tools. I use Claude, ChatGPT and Codex, GitHub Copilot, and models that run locally such as Qwen when a company's data must not leave its own servers. I do not sell licences and I am nobody's reseller: I choose according to the job, and if something better fits we change.
Where your data ends up. It is the first question everybody asks, and the answer depends on how the thing is built. When the material is confidential, the extraction runs on a model hosted on your own server or a European one, and nothing goes out to external services. When that is not necessary, the large providers' APIs are used, which cost less. The choice is made beforehand, together, and it is put in writing.
You can see it working elsewhere. In the management system I built for a digital agency, an assistant answers questions about the company's data instead of making somebody build a filter by hand. In GNAMMM the messages to the restaurant's customers go out by themselves over WhatsApp. These are not demonstrations: they are things running every day.
Not prototypes: systems in production, with measured numbers. Each case study sets out the problem it started from, how it was solved and how well it works.
It depends how many hours it takes away now, and that is the figure we start from. The first meeting is there to work out whether it is worth it: if the hours saved do not pay for the work I tell you straight away, because an automation that does not pay for itself damages both of us.
The first working piece in four to six weeks, because measuring accuracy means working on your real documents. Manifest reached 90.6% after several rounds on real emails, not on the first attempt.
Only if we decide that together. When the material is confidential I use models running on your own server or a European one, and nothing goes out to external services. When that is not necessary, the large providers' APIs are used, which cost less. The choice is made beforehand and put in writing.
It does get things wrong, and that is expected: which is why it proposes rather than decides. A person approves, and every correction becomes an example that improves the next pass. Accuracy is measured and visible, document by document.
No, and usually it is the opposite: the automation attaches to what you already have. Manifest puts the orders into the same system the client was using before, without anybody changing their habits.
Yes. My base is in Abruzzo, in central Italy, and I come in person where the distance allows it, but I have run these projects remotely too, for clients in other regions and abroad. For the initial half day, the one spent watching the process, I do recommend meeting face to face.
What Clients Say
‘The software Orion delivered went well beyond what I expected. He was precise, professional and followed my instructions to the letter. In short, if you are looking for software built around your needs, go to him, you will not be making a mistake.”
Fiverr review
‘’Thanks to Orion I got my first website and I am really pleased with the result, because it matches the idea I started with. Polite and professional, and always available for changes afterwards. I warmly recommend his work.”
Google review
‘’Orion built a store from scratch for my e-commerce in no time at all. I can only recommend him to anybody who wants their website done to a really professional standard. Very quick to reply, and faultless after-sales support.”
Google review
‘’An excellent professional, superb communication, attentive to every detail, and constant support. I recommend him to anybody who needs software built.”
Fiverr review
‘’Great seller highly recommend and provided a proffesional service as well as a polished project that met my expectation. Recommend this seller, very kind and great communication throughout the project!”
Fiverr review
‘’Perfect, excellent seller and an expert programmer, highly recommended. I will be building more software with him.”
Fiverr review
