Put your own data to work

and answers arrive fast

Most of the work is ordinary software: internal systems, data processing, process automation and connecting to what already runs in the company. I add AI where it has something to do, and I say so when the brief does not need it.

Yes, that is usually the heart of the job. I regularly integrate ERP, accounting and warehouse systems, SharePoint and DMS, Active Directory and SSO, and third-party APIs. Where no API exists, I design a safe way to get the data out without touching a system that works.

The application and its database can run in your own cloud tenant or on-premise, so documents and operational data stay with you. Where a language model is part of the solution, I call it through a commercial API (OpenAI, Anthropic, Google) in a mode where the provider does not use your content for training and does not retain it beyond serving the request, with EU regions and a data processing agreement. If the data must never leave your perimeter, we can talk about hosting a model of your own.

From sensors, PLCs and machines, or from the databases of systems you already run. I collect and normalise it into one place, which then feeds the visualisation, the history and the trends. Where a measurement is missing entirely, we add sensors only where the picture makes no sense without them.

Answers are grounded in specific passages from your documents and each one shows its source, so anyone can verify it in two clicks. When there is no supporting material, the assistant says so instead of inventing something. Before go-live we run a set of real questions from your company and score the accuracy with numbers, not impressions.

In most cases I call commercial APIs from OpenAI, Anthropic or Google. It is the fastest and cheapest route and it is enough for most companies. Where the brief demands it, I can host an open-source model such as Llama on your own or on dedicated infrastructure, and build a custom ML model or neural network. It is not the default: an own model means data to train on, hardware to run it and someone to keep it current. Whether it pays off in your case, I will say up front.

For internal systems it is usually just hosting and the database. An AI solution adds model usage on top, based on the number and size of queries. You get a concrete estimate for your volume as part of the audit, before anything is built. Once live, I track and report costs so they do not creep up unnoticed.

You do. Source code, documentation, infrastructure access and the processed knowledge base are handed over on completion. I do not work with vendor lock-in; your internal team or another supplier can take the project over at any point. I design solutions so that even the model can be replaced, because the best one today will not be the best one in a year.

I sign NDAs as standard, respect internal security policies and handle personal data in line with GDPR, including a processing agreement with the model provider where AI is involved. I work with your IT on deployment, access and operating rules, and I will supply whatever your security review needs from a supplier.

Then I will say so. That is exactly why we start with an audit rather than development. Sometimes changing a process or buying a module for a system you already own is cheaper and faster. Delivering something nobody uses six months later is a worse reference than a lost contract.

The audit takes days. A first pilot in real operation usually four to eight weeks from signing, depending on the state of your data.