May 6, 2026
  • capstone
  • nlp
  • client-project

Echtralex / Lexicography

A UMD capstone for a real client — automating a manual bilingual transcript-alignment workflow, with a measured ~100% speedup.

Overview

Full write-up in progress. Built with a team of 5 for CMSC435, in collaboration with Echtralex founder Michael Horlick — automating a fully manual bilingual (Chinese/English) transcript-alignment workflow that previously took 8–20 minutes per document by hand.

The client measured roughly a 100% speedup against the old workflow, and the system held ~98% alignment accuracy. Full story soon: the greedy matcher we threw out, the LaBSE + custom-scoring redesign that replaced it, and the HTML extraction work I owned directly.