Chrystian Schutz

Schutz’s CV describes a Frontend Tech Lead role at UBS, following engineering work at Collibra, HSBC and TietoEVRY. He studied information technology at WZIS University Copernicus in Wrocław. Alongside that work, he founded SafeLLM and created ThreadShelf. These projects address two practical problems in AI applications: controlling data and requests at a gateway, and retrieving useful material from scattered conversations.[2]

One searchable archive for AI conversations

ThreadShelf parses exports from several AI services into a common format, computes embeddings locally and stores the result in a searchable index. Users can find a conversation by meaning, reopen the complete thread and use the same archive through a web interface, command line or MCP server. Optional conversation generation can run through local llama.cpp or an explicitly external OpenRouter provider. The documentation distinguishes the local archive from that optional off-device generation.[4]

A poker experiment with paired deals

Schutz’s poker study compares language models, decision models and scripted baselines in heads-up no-limit Texas hold’em. It plays each deal twice with seats swapped and adjusts all-in outcomes using the players’ chances at that moment. These choices reduce the influence of lucky cards so the comparison better reflects decisions. The article reports model-specific results and uncertainty intervals. They support conclusions about this benchmark setup rather than general rankings of intelligence or financial skill.[3]

A local security gateway

SafeLLM applies several checks around model requests and outputs, including prompt-injection detection and sensitive-data controls. The project runs alongside Apache APISIX and offers configurable output handling. It illustrates Schutz’s interest in infrastructure that can be operated inside an organization’s own environment.[5]