Digitalization, AI agents and automation, built on a secure network and servers that keep running. And when your data shouldn't leave the company, the AI runs on your own hardware.
Working in IT infrastructure since 2020
Taking new projects
Six areas, one goal: less manual work and fewer surprises.
Focus
From paper and spreadsheets to systems that talk to each other.
We move recurring paperwork, spreadsheets and manual steps into proper tools: shared documents and approvals in Microsoft 365, forms that feed a database, and applications connected through their APIs, so nobody types the same data twice.
Learn moreFocus
AI assistants that work with your data, on your terms.
Chat assistants that answer from your documents, agents that draft replies, sort requests or prepare reports, and coding help for developers. They run on a private model on your hardware or on a cloud model you choose, with clear limits on what they can see and do.
Learn moreHelp for your team when something stops working: accounts, laptops, email and Microsoft 365. Remotely during working hours, and on site in Bucharest when it can't be fixed remotely.
Learn moreScheduled scripts and small services for imports, checks, reports and accounts, with logging and alerts.
Learn moreFirewall policies, Microsoft Defender, phishing simulations, email authentication (SPF, DKIM, DMARC) and backups that are tested.
Learn moreSegmented office networks, VLANs, VPN for remote staff, and a documented setup the next person can understand.
Learn moreThe firewall was set up years ago, rules have piled up and the person who did it has left. We find out what is actually there, remove what isn't needed, and split the network so one infected laptop can't reach everything.
They do it because it helps. We give them the same help on hardware you own: a private model with a chat window and an API for their editor. Nothing leaves your network.
We run realistic phishing simulations in Microsoft Defender, then short training for the people who clicked. You get the numbers before an attacker does.
Imports, checks, reports, new accounts. If it's repetitive and stable, we script it, schedule it and log it, with a human check where a mistake would be expensive.
Real projects, described the same way: what was wrong, what we did, what changed. Where we don't have a measured number yet, the gap is marked.
The team wanted an AI assistant for chat and coding, but company code and documents shouldn't go to a public service. No single workstation had enough GPU memory for a 27B model with a long context.
Load test with 10–16 people at once: 0 errors, about 22 tokens per second each, 262K-token context.
Read how we did itDiagram: users connect to Open WebUI or the API, which send requests to llama-server. llama-server spreads the model across four GPU workers over llama.cpp RPC: three NVIDIA and one AMD. Tokens are streamed back.
↑ answers streamed back
llama-server loads the 27B model and puts a part of it into each worker's GPU memory. For every token it coordinates the workers over the network. It runs under its own service account.
This splits one model across four machines; it isn't load balancing. Every request uses all four GPUs, because each holds part of the same model.
The firewall's data plane was running close to 100%, and nobody knew which traffic was causing it.
We found which traffic was filling the firewall and tightened the policies around it. Remote staff connect over VPN.
Nobody knew how the team would react to a convincing phishing email.
A measured picture of how the team reacts to phishing, from realistic campaigns instead of guesses.
Staff needed to run a handful of maintenance scripts. Giving everyone shell access to a server is how accidents happen.
People run the approved scripts themselves, and nobody needs shell access to the server.
Discounts in Romanian online shops change every day. Checking them by hand doesn't scale.
Offers are imported, checked and published by scheduled jobs, without manual work.
Recurring documents were put together by hand, and ad campaigns ran without reliable measurement.
Recurring documents come out of a script instead of being put together by hand, and campaigns run with conversion tracking.
We look at what you have: network, servers, accounts, backups. You get a short written report with the risks in order of importance.
We agree what gets done, in which order, for a fixed price. No open-ended hourly billing unless you prefer it.
We make the changes in small steps that can be undone. Mostly remote, at times that don't stop your team working.
You get documentation your next admin can actually use, and every password and account stays in your name.
Book a call, email us, or use the form. You get a reply from the engineer who would do the work.