Imported from tschomay/tschomay.github.io (AGENTS.md). Install upstream with npx skills add tschomay/tschomay.github.io. Copyright stays with the author.

Working on this site

This is Ted Schomay's personal site. It got a full rebuild in Aug 2026 after sitting mostly untouched since 2022 (see git log). Read this before editing copy, adding a Notes post, or restructuring anything — it captures decisions that were deliberate, not default, so they don't get silently undone by a future edit.

What this site is actually for

Ted isn't job hunting. The site isn't a resume mirror or a "digital business card" — that's just LinkedIn with extra steps, and LinkedIn already does static facts better. What a personal site can do that LinkedIn can't is show a personal touch: how Ted actually thinks, what he's like to build things with, and the judgment behind the work — not just the résumé line.

The thesis, stated on the homepage and meant to run through everything here:

What you do is only part of the picture — it's how you do it that has the bigger impact.

Concretely, that means the site is optimized to demonstrate connecting technical depth to business judgment, and personality/working style — not to list credentials. The homepage's "What I'm looking for and where I do my best" section exists for the same reason and is written in Ted's own words, first person, deliberately not third-party testimonials or pulled LinkedIn recommendations — he wants the site to say who he is, not advertise him.

Structure, and why it's shaped this way

• Nav is Home / About / Notes. Research (the PhD work) is deliberately not in the top nav — it's linked from About as a supporting deep-dive, not the lead identity, because Ted is a practicing industry data scientist now, not an academic. Don't promote Research back to the nav without a reason; it was demoted on purpose.
• No resume PDF. Removed deliberately — Ted doesn't want to maintain a synced PDF. Site directs interested parties to make contact instead. Don't re-add a resume link without being asked.
• Notes pulls itself. The homepage's "Recent notes" section and the /blog index both loop over site.posts via Liquid — publishing a new post is enough, nothing else needs manual updating.
• Visual design got its pass in Aug 2026. The site runs a custom system ("Signal & Noise" — built from Ted's own SVD/signal-extraction research as the visual metaphor, not a generic theme) instead of jekyll-theme-leap-day. See DESIGN_NOTES.md for the rationale, what's implemented, and the one thing still open (photo currency). The palette, type pairing, the notch motif, and the hero animation were deliberate choices approved by Ted — don't restyle incidentally while doing something else. If a future ask wants a different direction, treat it as its own design pass.

Writing Notes posts

The framing on /blog is: "less 'here's what I built,' more 'here's the non-obvious call I had to make and what it taught me.'" That's the default lens, not a mandatory formula — don't force every post into a business-judgment lesson if that's not actually what the project is about. That was a real correction from Ted on the first draft of the Auralux post: the initial draft picked two technical decisions and wrote them up like a capability showcase, and it read as manufactured insight bolted onto a project that wasn't actually about that. He wanted it cut down to what the project genuinely was — something built for fun, because he likes cool, colorful, glowy things — with one honest, specific thing he was proud of (the menu being part of the experience, not a chore panel), not a forced "and here's what this teaches us about engineering" angle.

So there are (at least) two legitimate shapes for a post, and the judgment call is picking the right one for the actual project rather than defaulting to one template:

1. "Look at this fun/cool thing I built." For personal/creative projects. The point is the idea and the fact that Ted spent real time on something whose only purpose is delight — not that it demonstrates a skill. Don't manufacture a lesson. If there's a genuinely interesting decision, mention it, but it's in service of the personality/craft angle, not a pretext for a business-judgment moral. Screenshots or visuals are a good fit here since the whole point is often "look at this."
2. "Here's the non-obvious call and what it taught me." For projects (work-adjacent or personal) where a real tradeoff was made — the kind of thing that shows how Ted thinks about ambiguous problems generally. This is the "connects technical detail to the big picture" mode from the site's core thesis. Structure: the problem in plain terms → the call that wasn't the obvious/default answer → what happened → a takeaway that generalizes past the specific project. There's a scaffold for this at _drafts/TEMPLATE.md — useful as a starting shape, not a checklist to fill mechanically.

There's also a category Ted mentioned but hasn't written yet: learning tools built for his own learning that others could use too — likely a third register, more teaching-oriented, not yet defined by an example. Use judgment if one comes up; probably closer to mode 2 than mode 1, but check in if unsure.

The "Philosophy" category

A third register, distinct from modes 1 and 2 above: posts that are explicitly about the "how matters more than what" thesis itself, told through a specific story Ted led (not a project he built solo). First one is _posts/2026-08-14-documentation-quest.md. These use topic: Philosophy in the front matter, which surfaces next to the date on /blog and the homepage's recent-notes list (see _layouts/post.html and blog/index.md — it's a plain string field, not a Jekyll categories/tags collection; don't build out a full category-archive system for this, the site is small and doesn't need one).

Use topic:, not category:, for this field. Jekyll treats category/ categories in post front matter as a reserved key that feeds the default permalink (/:categories/:year/:month/:day/:title), so a post with category: Philosophy silently gets a URL like /philosophy/2026/08/14/... instead of the flat /2026/08/14/... every other post uses. Confirmed by building locally — don't reintroduce category: without also deciding you want that URL-structure change on purpose.

The sharpest version of the thesis, distilled from the documentation-quest story: "how" is what turns a task people have to do into one they want to be part of — the test isn't whether the work got done, it's whether people who didn't have to be there started asking to join. Use that as the bar for whether a candidate story actually belongs in this category — it should have a concrete signal like that (people opting in, culture visibly shifting), not just "I did a project well and people liked it."

Ted's other recurring type of work: rescuing failed analytical products

Ted has mentioned, as context for future agents rather than as something to publish verbatim: a recurring pattern in the work he's drawn to is being brought into analytical/ML products that have already shipped and are failing in some way that isn't visible from outside — e.g. a proof-of-concept released as a real product with every customer on a diverged version and no scalable process behind it, or a legacy product so layered with departed people's work that it quietly stopped delivering its core output without anyone realizing. The technical fix (the ML/analytics piece) is often not the hard part; the hard part is restructuring everything around it — process, scope, stakeholder visibility — while keeping deliveries running and without alarming clients that major changes are underway.

Ted was explicit: don't put the specific case-study details of these projects on the site (employers, clients, the exact failure modes aren't his to publish). Keep it to the essence — that his "what" isn't just ML engineering or analytics, it's finding the gap nobody's named yet, bringing thought leadership to it, and connecting the technical detail back to the business need — the same connective work the site's core thesis is already about.

This has since been worked into the homepage: the Welcome section's second paragraph ("The work I'm proudest of, though, is usually the rescue...") now carries this, at Ted's explicit request (Aug 2026). Still no case-study specifics there — keep it that way. If asked to add more, resist naming employers/clients/products; add color to the essence (gap-finding, thought leadership, technical-to-business connection) instead.

Voice notes, regardless of mode:

• First person, honest, a little self-deprecating when true ("the implementation isn't doing anything special" is a real line from the Auralux post — don't be afraid to say a project isn't technically impressive if the point is elsewhere).
• No corporate speak, no manufactured stakes.
• Keep it to a handful of paragraphs. These are Notes, not essays.
• The title should sound like something a person would actually say, not a headline (compare "Auralux: I just like things that glow" to the discarded draft title "Auralux: the best UI decision was deleting a button" — the second sounds like a Medium post, the first sounds like Ted).

Before publishing

• Build locally to catch Liquid/markdown errors before pushing (see README.md/repo history for the Jekyll build workaround needed in some sandboxed environments — bundler can be broken there; installing the jekyll and jekyll-theme-leap-day gems directly and building without Bundler is the fallback).
• If real screenshots exist for a post, use them — but if you can't reliably get binary image data into a tool call intact (this bit us once with corrupted base64), don't guess-and-check silently. Ship the text, tell Ted what's missing and why, and let him decide whether to supply the image or wait.