Research automation
Turn messy research loops into inspectable scripts, receipts, dashboards, and human-in-the-loop workflows.
GitHub profile / work with me
I builds research automation and strange-useful knowledge systems.
This site is the public lab notebook: scientific ML notes, Sandy Chaos research architecture, Yggdrasil continuity work, and interactive tools for making hard-to-see structure easier to inspect.
If you are looking for the human-readable version, start here. If you are looking for help, the strongest fit is work that needs rigorous experimentation, automation, visualization, and honest uncertainty boundaries.
Current offers
Where I can help
Interactive lab websites
Build small static-first tools and public demos that help people understand a project by touching it.
Materials and ML bridges
Connect experimental materials data, crystal-structure representations, and graph-ML intuition into clearer working prototypes.
A bounded starting point
Research Automation Diagnostic - $125 pilot
Bring one manual, fragile, or confusing research workflow. We will map it in a 60-minute working session, identify the smallest useful automation or visualization, and finish with a short implementation map covering risks and next steps.
If the diagnostic reveals a clean build target, a fixed-scope prototype sprint can follow, typically starting at $500. No open-ended consulting commitment is required.
Proof of work
Useful places to inspect after GitHub
Materials Graph Lab
A toy crystal-to-graph lab for structural disorder, spin textures, and message-passing intuition.
Open app ->
Materials data bridges
A research note on connecting experimental materials databases to crystal-structure data for graph neural networks.
Read note ->
Yggdrasil
A continuity architecture for coordinating branching work, memory, trust, and temporal scale.
Read essay ->
Contact path
Start with GitHub, then email.
The GitHub profile is the main proof surface for now. For a concrete project, email what you are trying to build or understand, what already exists, what the deadline or constraint is, and what a useful first pass would prove.