Code and methodology belong together
A result is only credible when the hardware, model, input and measurement path are visible.
About geisten
Open engineering for Efficient AI.
geisten is the open engineering lab of Dr. Germar Schlegel. The lab optimises small models for maximum efficiency and equips them with capabilities they were never trained for — a ternary BitNet that outruns Microsoft's own runtime on a Raspberry Pi, and tiny models that drive a command line as agents. Every claim ships with a reproducible benchmark.
Working method
From system constraint to measurable implementation.
While much of the industry scales models up, we scale them down. The question behind every geisten project is simple: what is the smallest model that solves this problem?
We don't believe the future is one giant model running everywhere. It's thousands of focused models, each doing one job well, running where the data lives. That's why geisten starts from the constraints — memory, compute and the task itself — and publishes the measurements to prove what works.
Measurements over performance claims
Latency, throughput, memory and energy are properties of an entire system, not a model alone.
Local execution as the starting point
Privacy, offline capability, latency and access to hardware are engineering parameters from the beginning.
Research directions
Inference Runtime
A compact C23 inference engine for small CPU systems, with ARM NEON kernels and transparent performance measurements.
Tiny Models
Investigations into small language models, task fit, retrieval and evaluation on constrained hardware.
Local Agents
Autonomous agents with clear tools, narrow tasks and traceable evaluation criteria.
Embedded Intelligence
Models, computer vision and sensors for devices that need to work without permanent cloud access.
Who runs this
Dr. Germar Schlegel
Germar Schlegel holds a PhD in experimental physical chemistry and builds efficient AI systems in Düsseldorf. His work spans the geist inference engine, ternary model optimisation and agent harnesses for small models — all developed and measured in the open.
For press enquiries, talks or collaboration: g.schlegel@geisten.net
Community
Research is better in the open.
Discussions, bug reports and reproducible counter-measurements are welcome. Current artefacts, benchmark runs and notes are published on GitHub.