About Django.
I build AI software and co-founded Kamoo. Through Kamoo, I build applications for clients and provide training.
I'm Django de Vreng. I build production AI for clients: agents, MCP servers, on-prem deployments on a DGX Spark. I also train teams that want to build with AI themselves. I do that through Kamoo, where I'm a co-founder.
I mainly focus on agents, MCP, local models, evaluations, vLLM and on-prem setups, and the question of how you get this reliably into production. I want to know where a model is good enough, where an agent gets too much freedom, which evals are useful, and which architecture is still readable when you have to dive back in three months later.
I became particularly interested when language models could also use tools. They can retrieve documents and prepare a draft for someone to review, for example. I enjoy figuring out how to bring these steps together in software.
With an agent, I also look at what can go wrong. What data ends up in the logs? What happens when a tool does not respond? Can I trace why an answer was given? These questions come up as soon as other people use the software.
On-prem AI interests me because not all data can be sent to a hosted model. Some organisations handle personal data, internal documents or client data that needs to stay closer. I look at what must run locally and what an external service may process. I calculate whether running locally also makes financial sense on the page about what a DGX Spark costs per month.
On this blog I write about that work: projects and tests on the DGX Spark, including the decisions and problems I encounter. I collect the raw figures behind these articles in the arena.
Outside work, I exercise a lot and go hiking with friends a few times a year. It gets me outdoors and away from my phone for a while.