GRE Lexicon: a vocabulary trainer built for one demanding user

Preparing for the GRE, the bottleneck was not effort but triage. Twenty word lists disagree, several run past 4,000 entries, and none of them rank a word by anything you can inspect. The question worth answering was not what does this word mean, but which of these thousands actually deserve the next hour.

Sole designer and engineer, from data pipeline through interface.

• ▸Treated the corpus as the product: parsed and reconciled 20 sources, including three PDFs whose text layers were damaged, one of which had every space stripped out and needed dictionary-based reconstruction
• ▸Made importance falsifiable by scoring each word on how many independent lists carry it, then exposing that provenance on every card so the ranking can be argued with
• ▸Chose determinism over coverage for root families: a curated root table with an explicit blocklist, because a table that quietly files 'denigrate' under gratus is worse than no table
• ▸Designed mobile-first around a four-way swipe, then gave the laptop its own two-column layout rather than stretching the phone one

• ▸Swipe-first drill with scope filters, spaced repetition, undo, and cross-device sync with no account
• ▸Card backs carrying mnemonic, memory hook, root family, etymology, personal notes and full provenance
• ▸Search across the whole 9,917-word corpus that also matches on meaning, so 'stubbornly resistant' finds recalcitrant
• ▸A bring-your-own-key coach that marks your own sentence using the word, plus 564 embedded video explainers and Hindi on every card

• ▸A tool built to be used daily under real time pressure rather than demonstrated once, installable to a phone home screen and usable offline of any login
• ▸Deliberately local-first: no accounts, no backend, no data leaving the device, which also means a visitor can try the full drill in ten seconds