Every organisation I have worked with — hospitals, banks, universities, public institutions, start-ups — eventually faces the same question: who else can read this? For years the honest answer was “your cloud provider, and whoever they share it with”. We accepted it because the alternatives were hard.
They are not hard any more. Modern cryptography can be packaged so that a product team adds end-to-end encryption in days rather than years. Open models and commodity GPUs mean a capable AI system can run entirely inside the walls of the organisation that depends on it. The technology exists. What is missing is the default — the expectation that privacy is the starting point, not an upgrade.
Communication safety
Messages, files, meeting notes, medical records, contracts: this is the raw material of every institution, and it travels through more systems than anyone can list. End-to-end encryption is the only architecture where a breach of the server is not a breach of the people. With PrivMX we made that architecture reusable: developers get secure threads, stores and inboxes as building blocks, and never have to touch a key.
AI sovereignty
Generative AI is now being wired into those same data flows — summarising the records, answering questions about the contracts. Sending all of that to an external API is a data-governance decision disguised as a product feature. DeepFellow exists so that a regulated organisation can have modern AI on its own terms: on-premises or in a private cloud, with full control of data, cost and compliance.
Sovereignty is not isolation. It is being able to say yes to new technology because you know exactly where your data goes.
I came to this view from three directions at once: as an engineer who has built systems on both sides of the trust boundary, as a lecturer who has watched students inherit an internet that treats their data as inventory, and as a CTO responsible for the promises we make to customers. Those three perspectives agree. That is why I do this work.