Technology that transforms and educates.

A mid-size financial-services company needed its AI agents to stop being one-off software projects. Each new agent took a developer and a deploy, and its safety depended on who had built it.

We rebuilt the service as a platform where an agent is a document, not code, and where safety, limits, audit and human approval come built in. Six agents run on it in production today, across chat, voice and the company's app, one of them written by the client's own security team.

The platform above was built with our software factory: one engineer directing AI agents through a written method, with the client's engineers reviewing every change.

The agents build in parallel and check each other's work before a person sees it, and people keep every decision: what gets built, what is merged and what goes to production. In 37 days the factory completed more than four times the changes a team of four had completed in the 37 days before, with no rollback in production.

Every clinical trial is defined by a protocol, the document that describes how the study runs. A new global standard, ICH M11, changes how these documents are written. A top-30 New York-based biotech pharmaceutical company engaged us to build open-source AI tools that make the switch simpler, free for any team to use.

Medical writers upload an existing protocol and, within minutes, get it in the new format, with gaps flagged and every section traceable to the original. Instead of rewriting, they review.

It runs anywhere, including regulated pharmaceutical environments. And it shows that highly regulated work, like clinical trials, can be faster and easier to deliver than expected when the rules are well written, as they are in ICH M11.

An AI tutor that lets students from 6th to 9th grade practise one curriculum skill at a time, in conversation. It works from material written by their teacher and only asks questions: it never hands over the answer.

When the conversation ends, a second AI suggests a grade and shows the teacher where the student got stuck. The grade is the teacher's to give. The tool runs on our own agent platform, and we are refining it with the teachers of the Ninho Project's school before taking it to other schools.