- HTX — Custom software development with private AI | Trieste/
- Services: software development, production, security | HTX/
- From AI prototype to production software | HTX/
From AI prototype to production software | HTX
Take your prototype into production.
The prototype works on your computer. Then come the real users, the real data, security, the ERP to connect. We assess it, tell you what to keep and what to redo, and take it into production.
It works in the demo. Not yet in the company.
With Claude, ChatGPT or Cursor, anyone can now build a prototype in a few days. That is great news: it shows the idea holds up. But between a prototype that works and software the company can rely on, there are things you can't see in the demo:
- Who gets in, and to what. Login, roles, permissions.
- Real data. Volumes, edge cases, personal data, backups.
- Integrations. ERP or management system, CRM, email, archives.
- Errors. What happens when something goes wrong, and who notices.
- Dependencies. Outdated or vulnerable libraries, API keys in the code.
- Maintenance. Who updates it in six months, and how.
From assessment to production, in three steps.
Readiness assessment
We analyse code, dependencies, data, access and architecture. We deliver the real state of the prototype, the risks and a reasoned choice: reuse, refactoring or rewrite, part by part.
First paid phaseIndustrialisation
We put the plan into practice: authentication and permissions, error handling, integrations, automated tests, separate configurations for each environment.
Verifiable releasesAcceptance testing, release and management
Acceptance testing against the agreed criteria, release into production on your servers or a European cloud, then maintenance with activities written into the contract.
Documented checksA document that is useful even if you don't work with us.
The assessment is real work, not a disguised sales offer. We deliver it in a form you can use with your own team or with another supplier.
- State of the code: structure, quality, tests present and missing.
- Dependencies and access: libraries, external services, keys and secrets.
- Data: where it is, who sees it, what is needed to handle it correctly.
- Reasoned decision: reuse, refactoring or rewrite, component by component.
- Plan and estimate to reach production.
- Acceptance criteria: how we will know, together, that it is ready.
We work with AI, and we know where it goes wrong.
In the ASA-PS project, with Centro Ortopedico di Quadrante of the Ramsay Santé group, we compared 11 language models on standardised clinical cases. The study is published in Informatics in Medicine Unlocked (Elsevier). We know how to measure when a model is reliable and when it isn't: it is the same care we apply when assessing code that a model has written.
And we know how to bring software into real processes: for T&B Associati, a migration from Excel to the ERP went from 50 person-days to 1.5 days; for Podartis, 800 pages of PDF orders go into the ERP automatically, with anomalies flagged to an operator.
Before we start, people usually ask us:
The prototype was written with AI. Is that a problem?
No. We assess it like any other code. AI-generated code often works in the demo but neglects authentication, error handling, real data and updates: that is exactly what we check.
Our prototype uses an AI model. Will it have to use one in production too?
Not necessarily. Code written with AI and software that uses AI are two different things. If a model is genuinely needed, we assess where to run it, including privately on PRISMA; if a rule or an integration is enough, we tell you.
Will we have to throw everything away?
Rarely. The assessment ends with a reasoned choice, part by part: what to reuse as it is, what to refactor, what to rewrite. A complete rewrite is the last option.
Do we have to give you access to the code?
For a proper assessment, yes, read-only and after a confidentiality agreement. We don't ask for code or confidential documents in the contact form: those are shared later, through an agreed channel.
How much does the assessment cost?
The production-readiness assessment is paid technical work, with a scope and a price agreed beforehand. The first call, where we work out whether it makes sense to do one, is free.
Is security your main concern?
If the application is already in use and you need to show that it is secure, see software security. If you need to build from scratch, see custom software development.