I design frameworks other developers build on.
Three of these frameworks live in Python's desktop ecosystem: a theming and design-system layer, an icon system, and a full application framework on Tk. My day job is platform work within a supply-chain product, where the same instinct applies: build a cohesive solution and configuration framework that enables and expedites solution implementation. The connecting factor in all of these is API design.
Plenty of developers are satisfied with an API that works correctly; this is expected. But for me, it should also feel good to use — organized, discoverable, documented, easy to read. That is the difference between an API that is usable and one that is productive and pleasant to use. I spent most of my career as a consumer of productivity applications. The end user experience is always at the forefront of my mind when designing an API.
A theming and design-system layer for Python's tkinter. Thirty curated light and dark themes that switch at runtime, a single bootstyle keyword that styles any widget by intent, 2,000+ icons rendered from a font rather than shipped as image assets, and a declarative Theme() API for building a custom design system. Pure Python, one dependency.
primary.Striped.Horizontal.TProgressbar. The segments are order-dependent, largely undocumented, and — the real cost — a wrong one fails silently. You get an unstyled widget and no error, so the feedback loop is guesswork. Tkinter does not ship with icons. You must use png images, which is a packaging nightmare.bootstyle accepts a fixed keyword vocabulary in any order, resolves it to the underlying ttk style name, and raises on anything it doesn't recognize. Colors come from whichever theme is loaded rather than from literals at the call site, which is what makes runtime theme switching possible at all. Finally, icons are built-in and were added through a created technique I deployed on the `tkinter-icons` project by creating on-demand images through a bootstrap font icon. This gives you access to thousands of customizable icons.# ordered, undocumented, # silent when wrong ttk.Progressbar( style="primary.Striped.Horizontal.TProgressbar" )
# any order, validated; # a typo raises ttk.Progressbar( bootstyle="primary striped" )
The same keyword vocabulary styles every widget in the library. Change the theme at runtime and all of them follow:
The scheduling core of a medical practice, built as a Postgres project: what does it take to make double-booking structurally impossible rather than merely unlikely?
slot_taken error.The constraint — migrations/0003_occupancy.sql, trimmed
CREATE TYPE claim_kind AS ENUM ('appointment', 'hold', 'time_off', 'admin'); CREATE TABLE time_claim ( id uuid PRIMARY KEY DEFAULT uuidv7(), provider_id uuid NOT NULL REFERENCES provider(id), during tstzrange NOT NULL, kind claim_kind NOT NULL, expires_at timestamptz, -- holds only: NULL = never lapses released_at timestamptz, -- set on cancel; the row stays -- two live claims on one provider cannot overlap. -- not a check the app runs. a fact about the table. CONSTRAINT no_overlap EXCLUDE USING gist ( provider_id WITH =, during WITH && ) WHERE (released_at IS NULL) );
Cancellation is a release, not a delete. Appointment status changes cascade into released_at through a trigger — canceled and bumped free the time, a no-show deliberately does not — and the row stays, so the history survives. The same table carries time-limited holds, which is the groundwork for an ASAP waitlist: a released slot is held for one patient at a time, and a hold that nobody takes lapses on its own rather than needing a sweeper to be running.
A full application framework on Tk, and the largest of the three. Declarative context-manager layout where the code hierarchy is the UI hierarchy; reactive signals for state that flows between widgets; an event stream API with composable filtering, debounce and throttle; hot reload that keeps your place across saves; and a CLI that scaffolds a project, runs it, and packages it to a standalone executable.
A font-based icon system for tkinter — Material, Font Awesome, Lucide, Bootstrap, Simple Icons and eleven more behind one uniform API. Shipped as eighteen distributions that install as extras, so you pick the families you want and the distribution names never surface in your code.
A companion PWA for the Catan board game: Firebase for authentication and remote storage, Dexie and IndexedDB for local persistence so it keeps working offline. Source.
A fast, installable book reader for the memoirs of Duane H. Klepel and related works. One layout that reflows from a two-pane desktop view to a single phone column, with bookmarks and reading position carried across all 165 pages. Source.
I find the gaps in the platform, propose what should fill them, and then build it. Shared capabilities — Workflow Builder, Dataset Builder, Parameter Guidance, configurable web forms — designed so implementation teams can adapt them without pulling in engineering. Underneath that, the high-volume operational workflows themselves, including a re-architecture of core task management for the batch-driven operating model of a Fortune 100 retail enterprise.
Predictive modeling, sentiment analysis, process automation, and intelligence gathering through document and web-based extraction — delivered for internal stakeholders and owned end to end, from requirements discovery through knowledge transfer.
I came into engineering from finance and data analysis: five years of budgeting, forecasting and pricing work at Time Warner Cable and Grant Thornton, a CPA license, an M.S. in Accounting, and five years teaching graduate and undergraduate accounting courses as adjunct faculty at Indiana Wesleyan University.
My experience in various fields and domains has proven to be a practical advantage as I can more easily connect the product and design goals to the needs of the customer and end user. Hard problems are often operational and data-shaped rather than purely technical. I'm most useful where the right answer isn't obvious yet and someone has to weigh tradeoffs across the whole system before committing.