When a Good Idea Becomes Production Software
How a bottom-up idea from one employee turned into production-grade software — running on a local AI on EFINITI hardware in a German data centre.
A normal week in a HILO office
There is a moment in every office when a document in your hand stops being routine and turns into a question: how many more times do we have to do this by hand before we build a solution for it ourselves?
At the HILO office on the Lower Rhine, that question became concrete. Incoming PDFs, extracting amounts, sorting them into the correct category — repetitive work, error-prone, the same loop every day, labour-intensive and expensive.
Catrin, working in client services and with a strong technical inclination, saw the lever: this could be done by a program.
With an AI code assistant, a first prototype was built — faster than any classical development cycle would have allowed. The owner, Thomas, planned the deployment where the rest of the office IT had been living for years: on a server hosted by EFINITI Services. Before go-live, the EFINITI team performed a standard step: code review.
The turning point
The review surfaced two things Catrin could not assess on her own: the chosen AI model was not suitable for production use, and several security gaps were lurking in the code. This is not a criticism of the developer — this is the point at which initiative meets operational reality.
Instead of taking the risk into hosting, EFINITI proposed a different path: migration into a Pandora workspace, switching the code generator to Claude Code, and moving the AI analysis itself onto local EFINITI hardware with a local AI inside a German data centre.
What felt different
With Pandora, the work rhythm changed. Every code change was verifiable in the browser within seconds via the Showcase Preview — iteration time went from days to hours. The code lived safely in the EFINITI repository. The build and deploy pipeline ran directly out of Pandora. Maintenance is handled by Pandora agents.
“When I started, my biggest worry was: is this really safe, or are we building ourselves a problem? EFINITI reviewed the code, told me honestly what wouldn't hold up, and showed me how we could realise the same idea robustly with Pandora. I could continue with my draft instead of starting from scratch — for me that was the decisive difference. And the Showcase Preview did the rest: I see every change in the browser immediately. That cut our iteration from days to hours.”
— Catrin, Client Services & Development
The point: the AI stays in-house
The actual document analysis happens on a GPU in EFINITI's hands. No cloud call, no US provider, no third-country transfer. Documents are analysed, amounts are extracted, sorting is rule-based — all within German jurisdiction. Fully GDPR-compliant by design, not by after-the-fact reassurance.
A lean web app with user login serves the internal use case. Each unit sees only what it is allowed to see. Internal use only. Pandora customers pay a platform fee for Pandora usage. Looking ahead, token-consumption-based billing for Pandora inference customers will be added, as well as an optional MSP fee for extended maintenance.
What this means for the owner
“What convinced us: we could continue seamlessly where we had left off. Nothing had to be rethought, no large adjustments, no team retraining — our AI subscriptions were simply moved into Pandora. What did change: we no longer have to worry about infrastructure, security and compliance ourselves. And if something does come up — EFINITI has been our MSP for years, they know our infrastructure and they're ready to step in. For an office of our size that is, frankly, next level.”
— Thomas, Owner & Managing Director
The building blocks that made this possible
Provider-agnostic code generation
Optimal code quality and security baseline — without additional IT staff.
Showcase with feedback picker
Every iteration visible in the customer's browser within seconds.
Live deployment in a DE data centre
Hosting under German jurisdiction — without the office having to hold server know-how itself.
Local inference on enterprise GPUs
AI computation physically in Germany — without cloud risk.
Pandora agent maintenance
App care, logs, updates — automated by the workforce.
Tenant separation by design
Tenant, tier model, audit trail, billing — all in one console at EFINITI.
What this story shows
Today, ideas can come from the middle of an office, not just from the IT department. AI code assistants make that possible. What they cannot do: guarantee that an idea will become production-ready — safe, compliant, economically viable.
This is exactly where EFINITI's Pandora comes in. We take the draft an employee or an owner has built — and carry it through: with the platform, with hardware, with security review, with maintenance. One contact, many supporting pillars, a relationship of trust that has grown over time.