Independent engineering lab · kullhem.io
Kullhem Labs is an independent engineering lab for practical software, automation, infrastructure, hardware-adjacent tools, and experimental interfaces. Rough ideas become working systems here: dashboards, bots, integrations, edge services, utilities, diagnostics, and prototypes that start as hacks but are built with real engineering discipline.
Rough ideas becoming working systems.
Local models on real hardware. Multi-agent engines, routers, benchmarks, and the tooling to find out what actually runs in parallel.
Small systems that reduce repetitive work, connect services, and make everyday technical workflows easier to manage.
Bots, bridges, notification tools, and interfaces that help information move cleanly between people, apps, and devices.
Services designed for small servers, Raspberry Pi systems, Docker, local networks, and practical self-hosted environments.
Electronics notes, device interfaces, diagnostics, and software that connects digital systems to the physical world.
Dashboards, event pipelines, logs, metrics, and debugging utilities for understanding what systems are actually doing.
Projects that do not fit neatly into a product category, but are useful enough to build anyway.
Spotlight · Hardware-adjacent experiments
A keyboard, mouse, and clipboard bridge where the second computer sees real USB hardware.
Slide your cursor off the edge of a Linux desktop and it comes out on the machine next to it, carried by a Raspberry Pi 4 in USB gadget mode. The controlled machine needs nothing installed, no network path back to you, and it works at the login screen, elevated prompts, and installers, because it is just a keyboard and mouse as far as the host knows.
Stop putting every model behind a help desk. Give them a room, a game, and each other.
An AI text RPG driven by a parallel multi-agent engine. Several local LLMs each take a seat at the table, a Dungeon Master, a Narrator, and a party of Players, and play an adventure seeded from a single topic. The engine owns the ledger (dice, HP, inventory, map); the models own the story. Watch, whisper into the world, or pull up a chair yourself.
Benchmarking parallel subagent collaboration between two local LLMs across a three-GPU box.
A research workspace for finding the best way to run Qwen and Gemma at the same time on one machine: subagents, a proxy-based router, or a hybrid. Includes a GPU monitor dashboard, ten subagent definitions, benchmark logs, and written findings on where parallelism is real and where llama-server quietly serializes it.
Big Friendly GPU POD: Dockerized Stable Video Diffusion with TensorRT on an RTX 3090.
A complete Docker environment for running Stable Video Diffusion image-to-video XT 1.1 with TensorRT optimization. Automated model download, engine building, and a repeatable setup tuned for a 24 GB consumer GPU.
Engineering philosophy
Common tools and patterns
Status
Many repositories here are experiments, prototypes, or active workbenches. Some are polished. Some are intentionally messy while the idea is still being discovered. That is part of the lab.
This site is a static Astro build on GitHub Pages. Nothing here needs a backend, and the source is public.
