Patient Intake
Design Lady LLC, independent, 2026, language technology. Physical dictionaries for Dominican Creole exist. No digital tool did — full stop, not "a bad one existed," none.
Diagnosis
A gap, not a redesign. The actual diagnosis was less "what's wrong with the current tool" and more "why has no one built this," which is its own kind of design problem: proving something is buildable fast enough that the absence stops being a reasonable excuse.
Prescribing Physician
Sole designer, sole builder, sole everything. Concept, data pipeline, interface, and store listing, sitting from empty folder to live install in a single sitting.
Treatment Protocol
- Built an OCR pipeline to extract entries from the physical dictionary source material, since no digital dataset existed to start from
- Used Claude and Gemini in combination to structure and verify the extracted entries against each other, catching OCR errors neither model alone would have flagged
- Designed the extension interface last, once the data itself was trustworthy, deliberately not designing the UI first and hoping the data would cooperate
Dosage Limits
Self-imposed timeline: under 3 hours, concept to live listing, as a deliberate constraint to prove the build pattern (solo, AI-assisted, same-day) rather than an externally imposed deadline. Deprioritized: audio pronunciation for every entry, scoped out of v1 to hit that window, flagged as the clear v2 priority.
Clinical Trial Artifacts
The data pipeline that turned scanned dictionary pages into verified, structured entries.
The shipped extension itself: the onboarding/about screen plus all five main screens.
Efficacy Results
8,550 dictionary entries, live on both Chrome and Firefox shipped in under 3 hours.
Adverse Reactions Disclosed
The under-3-hours constraint, while a strong proof point, meant zero user testing before launch. Real usage since has surfaced a handful of entry-formatting inconsistencies from the OCR pass that a testing round would have caught, now queued for a cleanup pass.