Building Zyrabit / 01
Practical AI, close to the people using it.
I work across software, infrastructure, and applied AI — exploring how small, efficient models can make useful technology easier to run, understand, and own.
I am currently building Zyrabit: products, interfaces, and teams around a simple idea — useful AI should be accessible, local, and owned by the people who depend on it.
The idea
AI becomes more useful when it can run close to the data.
01Small models can be enough for focused tasks.
02Local execution can reduce dependency on external APIs.
03Open source makes systems easier to inspect and adapt.
04Real adoption depends on hardware, documentation, and examples.
Current project
Zyrabit — local AI infrastructure for real hardware.
Zyrabit is an open-source beta/MVP exploring how developers and organizations can run offline, traceable AI systems using infrastructure they already have.
Explore Zyrabit ↗A working example
Local inference on a Mac.
A measured comparison on Apple M1 Pro hardware with 16 GB of unified memory. Same question, two quantized models, different trade-offs.
“What are the three primary colors?”
Qwen responded almost twice as fast in this test because it does not stream internal reasoning tokens before the final answer. Results are tied to this exact hardware, runtime, prompt, and model build.
Where this can matter
One infrastructure layer, many contexts.
These are possible adaptation contexts, not customer cases, certifications, or production claims.
Private document search and clinical knowledge assistance.
Internal policy and procedure retrieval.
Controlled deployments in private or disconnected environments.
Local access to manuals and operational knowledge.
Reproducible experiments close to the data.
Personal context
The work is technical. The way I arrive at ideas is human.
Coffee, music, travel, photography, interfaces, and long walks all shape how I think about systems: through rhythm, constraints, repetition, and small details.
About me ↗