A terminal RPG where an LLM acts as dungeon master: narrating scenes, offering choices, resolving actions, and rolling dice against a simple d6 rules system, all inside one TUI app. Every turn gets appended to a story book, producing a readable narrative log of the whole campaign.

Built to run against a local LLM on a personal GPU as well as hosted APIs (OpenAI, Anthropic, Groq, etc. via litellm), with structural guardrails — a strict per-turn output format plus a second, separate LLM call that validates each action before it’s narrated — needed to keep smaller local models playable.

Status: Finished.

Project page: https://git.dejvino.cz/dejvino/splinter-keep

Running HTTP and HTTPS on the same port with Apache. They said it couldn’t be done. They were wrong!

https://github.com/Dejvino/https-multiplexer

I’ve modified a simple Python port forwarding utility to act as a port multiplexer that can automatically forward HTTP and HTTPS requests to the appropriate ports. If the request looks like an HTTP in plain text, it forwards it to port A. Otherwise it is assumed to be HTTPS and is forwarded to port B.

Now you can run your web applications from a single port, regardless of using HTTP or HTTPS. Hooray!

My Master Thesis software suite. The aim was to create a high-performance correlation computation software + visualisation framework for large datasets.

vizualizaceMultipleSignalsMap

Corrputer

Correlation computation software. Written in C++ with focus on high performance, uses MPI for running on any kind of setup, from one to many computers.

Source code: git.dejvino.cz/masterserver

Visputer

Data visualisation framework in Python built on top of PyQtGraph. It interconnects with Corrputer quite nicely.

Source code: git.dejvino.cz/mastervis