VidScope

Grounding strictly in the transcript is the right call for the hallucination problem, but it makes me wonder about the opposite failure: the tools that never get said out loud. In conference talks the library name is very often only on the slide or in the code sample, while the speaker says "this one" or "the usual suspects". And the names that are spoken tend to be the ones ASR mangles - Nuxt and Next, Pinia, Vite, anything that is not a dictionary word. Do you OCR the frames as well, or fuzzy-match spoken candidates against a known package index before you trust them? That is the gap I would check first on a 45-minute talk.

Hi, maker here. I built VidScope after spending way too many evenings pausing and rewinding conference talks just to catch the name of a tool someone mentioned once. Most summarizers give you a vague recap; VidScope is built specifically to extract concrete technical references (tools, frameworks, libraries) with their official links, strictly from what's actually said in the transcript. Happy to answer questions about the 0% hallucination claim and how it's enforced.