Lecture Lens — live glossary for your lectures

You are studying in a second language. The lecturer says a word you have never met, and by the time you have decided whether it mattered, three sentences have passed. Subtitles don't help, because you heard the words. Translation doesn't help, because the sentence wasn't the problem. Lecture Lens follows the lecture with you. The browser streams the microphone straight to AssemblyAI's Universal-3.5 Pro, and the transcript comes back in about half a second. Every word the course has never used passes two cheap filters and then a judge on AssemblyAI's LLM Gateway, which reads the word together with the sentence it was said in and decides whether it is a term of this subject. If it is, the word is highlighted with a one-line definition and added to today's glossary. The course remembers it, so it is never "new" again, and the known terms go back to AssemblyAI as keyterms, which is how "scanf" stopped being heard as "scanner". If you lose the thread, the "I'm lost" button explains how the new terms of the last three minutes connect. Everything is measured on real lectures. Letting the judge read the sentence cut false highlights on the benchmark from 16.3 of 19 to 3.0. On a live 16-minute lecture it caught 18 of the 24 terms counted by hand. 160 automated tests cover the server and the page in a real browser; for the key ones the bug was put back to check that the test fails. The README lists every limitation with its measured size. Why it doesn't talk back: it runs in a lecture hall. A voice agent that speaks over the lecturer would cost the student exactly the sentences it is meant to save.