Hardware, models, and inference.
Guides and interactive explanations for choosing hardware, understanding model architecture, and running models locally.
Background & Method
Researching technical systems from first principles.
I research local AI infrastructure, systematic trading, and technical systems. The work combines practical experiments, benchmarks, software, and clear explanations of how the underlying systems behave.
Systematic trading evidence.
Reproducible experiments on market regimes, filters, execution, and where intuitive ideas fail against data.
Tools that make research repeatable.
Local-first software for collecting evidence, testing hypotheses, and keeping technical decisions inspectable.
The person behind the notebook
Self-taught, and learning as the work requires.
I am self-taught. There was no formal path into software or AI — it came from independent study, reading, and building things until they worked. That background shapes the site: the writing starts from how systems actually behave, because that is how I learned them.
I also pick up practical skills as the work needs them. When a hardware question cannot be answered from a spec sheet, that includes electronics and soldering — enough to open the case, test the board, and find out what the numbers are doing.
The through-line is not a domain. It is the method: learn systems from first principles, build things rather than collect opinions, and let the measurements decide what is true. The boundaries between local AI, trading research, and software are arbitrary; the way of working does not change when the subject does.
No heroic origin story — just a practice, published here as it happens.
Workbench
How the work actually gets done.
The site is grounded in practical technical investigation — running things locally, measuring them, and writing down what held up.
Experiments before claims.
Each piece of writing follows the same loop: run the model or idea locally, test the hardware and runtimes, benchmark where it matters, build the tooling that keeps it repeatable, and document the tradeoffs that come out the other side.
Threadripper Pro, dual RTX 3090.
- AMD Threadripper PRO 3945WX
- 2 × NVIDIA RTX 3090, 24 GB each
- Debian Server
This is the machine behind the local AI guides. Read the complete workstation build, the RTX 3090 benchmarks, or the multi-GPU motherboard guide.
2 × NVIDIA DGX Spark.
Two NVIDIA DGX Spark systems are incoming. They are not part of the tested environment yet, so no benchmarks or hands-on results from them appear on the site.
Existing coverage of the platform, based on published data, lives in the DGX Spark guide.