How to search through 50 YouTube dev talks without rewatching a single one

by

I bookmark conference talks the way I used to bookmark Stack Overflow answers: aggressively, then never revisit them. My watch-later playlist hit 60 videos around March, and I knew about 40% of the content was things I’d already half-learned from skimming. The problem wasn’t finding the talk — it was finding the part of the talk.

The symptom: I remembered a conference talk where the speaker explained something about nested routing that I wanted to reference. Twenty minutes into rewatching, I realized the specific moment I needed was in a different talk entirely. Two hours gone.

The search I finally pasted into Google: search YouTube video transcript by keyword without downloading.

Here’s the fix I landed on.

The wrong paths I tried first

YouTube’s native transcript (three-dot menu → “Open transcript”) is the obvious first move. It works for a single video, but it doesn’t cross-search. You’d have to open each talk, Ctrl+F the transcript, close it, open the next. For 60 talks? Not happening.

I also tried yt-dlp --write-auto-sub to download VTT subtitle files and grep through them locally. It works, but VTT format is noisy — timestamps repeat every three words, lines duplicate from the rolling captions. You need a cleanup script before it’s searchable, and that cleanup script becomes a small project. I got halfway through writing it before I found a better path.

# VTT noise you'd have to strip — not worth it
00:04:12.340 --> 00:04:14.780
nested routing nested routing
00:04:13.500 --> 00:04:16.100
nested routing is handled

The actual fix: utubetalk

utubetalk.com converts any YouTube video into a full, searchable transcript in about 10 seconds. You paste the URL to their Telegram bot, it returns the full transcript, and the video gets indexed into your personal library at utubetalk.com/my.

No install. No account required to try it. $5/month for unlimited videos once you’re past the free tier.

The workflow

  1. Find a YouTube talk I want to index. Paste the URL to the Telegram bot.
  2. It returns the full transcript in ~10 seconds. I don’t read it — I just let it get indexed.
  3. When I need to find something, I go to utubetalk.com/my and search a keyword across my entire library.
  4. It returns matched excerpts with timestamps. Click the timestamp, you’re at the exact moment in the video.

For the 60-video backlog: I batch-submitted them in one sitting. About 15 minutes total.

Where this specifically helps dev workflows

  • Conference talks (PyCon, JSConf, GOTO, Strange Loop): architectural decision talks where someone explains “why we built X this way” — the kind of content you want to cite 18 months later when you’re making a similar call.
  • Long tutorial series: “He mentioned something about the context API in part 7 of this 12-part series.” Searchable now.
  • Migration guide videos: Framework authors often demo migrations on YouTube. The transcript means you can Ctrl+F for the exact error message they show on screen.
  • Architecture walkthroughs: When someone talks through a codebase live, the commentary is where the value is. Indexed.

Why I didn’t just build this myself

I briefly considered it: a Python script hitting the YouTube Data API for captions, storing in SQLite, exposing a small search endpoint. The YouTube captions API requires OAuth for non-auto-generated tracks, and many creator-uploaded subs aren’t accessible via the API at all — only the auto-generated ones are reliably available. You’d need fallback logic for both caption sources, plus the timestamp alignment logic for the rolling-caption dedup problem from the VTT format.

I estimated 8 hours to do it right. At $5/month, the break-even is 19 months. I subscribed.

Edge cases / gotchas

  • Videos without captions: If a video has no auto-generated captions and the creator hasn’t uploaded subs, it can’t be indexed. Private and unlisted videos won’t work either.
  • Transcript accuracy: Auto-generated YouTube captions are ~90–95% accurate on clear speech. Technical library names with unusual phonetics (Kubernetes, Svelte in early auto-cap days, Zig) sometimes get mangled. Use it to find the timestamp, then listen for the actual quote.
  • Free tier cap: I hit it indexing the 60-video backlog. The $5/month plan is unlimited.

TL;DR fix

Share your YouTube talk URL to the Telegram bot at utubetalk.com, then search your entire video library at utubetalk.com/my — full transcripts, word-level search, timestamps back to the exact moment.

Leave a Comment