The homepage is the scrapbook — a snapshot of who I am and what I’m building right now. This is the work-log — longer-form notes, the why behind specific decisions, working code, half-finished thoughts I want to come back to.

A few topics I expect to write about:

• the actual benchmark methodology behind the 8× phase-field speedup
• what worked and what didn’t when wiring SynaCAD’s solver layer to the LLM
• short Bruhn / Niu / ESDU walk-throughs as I revisit them
• occasional notes on Hindustani classical music, when a raga and a problem rhyme

Posts go in _posts/YYYY-MM-DD-slug.md. Markdown, Jekyll handles the build.

Why “Synapse”?

A synapse is the place where a signal becomes a decision. A neuron alone doesn’t think; the connection does. SynaCAD (“Synapse CAD”) is the connection between two things that have rarely talked to each other: the classical mechanics canon (Bruhn, Niu, ESDU, Lekhnitskii — a hundred years of structural reasoning) and modern generative models (LLMs that can read your sketch, your photo, your sentence).

It is not “ChatGPT for CAD.” It is a working engineer with a textbook open on its lap.

The vision

Anyone, anywhere, designing real engineered components from a sentence — and getting back a part that a real machine shop can hold tolerances on, with citations from real handbooks for every number.

What it actually does

The non-negotiable

Every number SynaCAD outputs is computed by a validated solver. Not invented by the language model, not interpolated from training data, not vibes. The LLM understands your intent, picks the right classical method, and writes the report. The numbers come from a function that, given the same input, returns the same output — and whose source you can read.

This is the line between “interesting demo” and “thing you can put your name on a drawing for.”

Design decisions worth naming

1. Solver-first, LLM-second. The model decides which calculation to run; the calculation runs in plain Python. No tool call, no hallucinated stress.
2. Citations for every number. Every result links back to a page in Bruhn / Niu / ESDU or to a peer-reviewed paper. If we can’t cite it, we don’t ship it.
3. Open by default. The bolted-joint solver is already MIT-licensed. The phase-field solver goes public the moment the paper hits arXiv. SynaCAD’s classical-tools layer follows.
4. Manufacturing-aware from day one. Drawings come out with GD&T that’s actually holdable on a real shop floor. We are not trying to impress a CAD reviewer; we are trying to make a real part.
5. No mystery in the loop. A junior engineer should be able to read SynaCAD’s report and learn why the part is sized the way it is, not just trust the answer.

Where we’re at

In active development. Currently pitching for the O’Shaughnessy Fellowship 2026. Public alpha planned for 2026.

If this resonates — collaborators, manufacturers, advisors, students who want to test it on their own parts — get in touch.

More about SynaCAD → the project site.