In this video I challenged myself to build an AI startup MVP in 24 hours—mainly to prove it’s actually possible if you stop overthinking and just ship. I started by brainstorming a bunch of ideas (Chrome extension autocomplete, doc-search chatbot, AI image stuff), but the reality is the big players already dominate those spaces. So I pivoted hard and went back to a real problem I’d had recently: getting stuck in a long tutorial and wanting a faster way to search a video for the exact answer. The MVP I built pulls a YouTube transcript, stores it, and lets you chat with it. I tested a few transcript libraries, landed on one that actually worked reliably, then wired the transcript into the OpenAI ChatGPT API to answer questions like “what is this video about?” From there I added a vector database using Astra DB (DataStax) so I could store the transcript plus embeddings and reuse existing videos without reprocessing everything. On the front end I kept it simple: an index.html UI with Tailwind, and a lightweight Node.js backend that takes a URL + messages, talks to Astra + OpenAI, and returns responses. It’s not perfect (long transcripts hit limits), but it’s a working MVP—and the next step is chunking transcripts and retrieving only the relevant sections.