learnaiwithrafa
Multi-ToolWorkflows

Turn Any YouTube Talk Into Your Personal AI Tutor

You forget most of what you watch. Here's a 4-step workflow to turn any conference talk or course into an AI tutor that quizzes you — with a copy-paste prompt.

6 min read4 sources
  • #notebooklm
  • #gemini
  • #learning
  • #upskilling

Here's the uncomfortable part of leveling up: I've watched hours of genuinely great conference talks — consensus protocols, staff-engineer career arcs, the lot — and a week later I couldn't reconstruct the core argument of a single one. Watching is passive, and passive doesn't stick. That's not a motivational cliché, it's measured: Freeman's team pooled 225 studies across science and engineering courses and found that students who actively work through material score about 6% higher on exams (a jump from the 50th to the 68th percentile) and fail at roughly half the rate of the ones who just sit and absorb a lecture. A YouTube tutorial is the same trap with a better thumbnail.

The fix isn't watching it twice. It's making the video talk back — explain itself, question you, and grade you. Two free Google tools do exactly that in about five minutes: NotebookLM turns the video into a knowledge base, then Gemini Canvas turns that knowledge base into an interactive tutor app. As an EM interviewing for international roles, that's the gap between "I watched a talk on distributed systems" and being able to reason about one out loud when an interviewer pushes back.

The 4-step workflow

1. Pick a source worth the effort

Grab the URL of something you actually want to own. A single video works; a full playlist works better, because the richer the source, the sharper the tutor. The ones that pay off most for engineers:

  • a system-design interview course — ten-plus videos of exactly the patterns you'll be grilled on
  • a conference track on distributed systems, databases, or reliability
  • a "build X in a weekend" series for the framework you keep avoiding
  • a staff/principal engineer talk on scope, influence, and technical strategy
  • a deep-dive playlist on Kafka, Postgres internals, or whatever last bit you in prod

One NotebookLM notebook holds up to 50 sources and 500,000 words on the free tier, so an entire talk series lives in one place instead of scattered browser tabs.

2. Feed it to NotebookLM

Go to notebooklm.google.com, create a notebook, click Add Source, pick YouTube, and paste the link. NotebookLM transcribes the whole video and turns it into a searchable, citable knowledge base you can interrogate — every answer points back to the exact spot in the source it came from. Right away you can ask it questions, have it draft a study guide, or generate an audio overview (a podcast-style walkthrough of the material) for the commute.

3. Generate the study layer

The strongest in-notebook output is a Video Overview: narrated slides that pull the actual diagrams, quotes, and numbers out of your source rather than hand-waving a generic recap — Google built it for exactly the "explain the data, make the abstract concept tangible" job. You can aim it at your level too ("assume I know nothing" vs. "I'm senior, skip the basics"). These are handy on their own, but the free tier caps you at daily limits of 3 audio/video generations and 50 chat queries, so spend them on the talks that matter — the study layer is really the warm-up for step 4.

4. Build the interactive tutor in Gemini Canvas

Open Gemini, click Add files, and attach your NotebookLM notebook — now Gemini has the full transcript and structure of your video. Then switch on Canvas, Gemini's app-builder. Instead of returning a wall of text, it writes and live-previews a working HTML/React app right in the chat, auto-saving the code as it goes and even surfacing a console if something breaks. Paste the prompt below with Canvas on and your notebook attached, and you get a real quiz app back — not notes about one.

The reusable prompt

Paste this into Gemini with Canvas on and your notebook attached:

Build an interactive study app in Canvas from the attached notebook.

The app must:
1. Open with a 30-second plain-English summary of the topic.
2. Walk me through the key concepts one at a time. For each: a short
   explanation, one real-world engineering example, and a
   check-your-understanding question I must answer before continuing.
3. Include an "ask anything" box that answers ONLY from the notebook
   and cites which part of the source it used.
4. End with a 10-question quiz (mix of multiple choice and short answer)
   with feedback after each question.
5. Score me at the end and list the exact concepts I should rewatch,
   based on what I got wrong, with timestamps.

Keep the interface clean, mobile-friendly, and colorful with clear
hierarchy — a tutor app, not a textbook. Make it feel good to use.

Every clause is doing work. The per-concept check question forces recall before you move on; the grounded ask-anything box stops the tutor inventing answers; the quiz plus targeted rewatch list turns your mistakes into a study plan instead of a grade. Drop any one and you're back to a fancier way of reading.

Share it — build once, everyone learns

Canvas apps share through a public link. Build the tutor once and send the link to your team, your study group, or the new hire ramping on the same system — same notebook, same tutor, zero extra work for you. I keep a distributed-systems tutor bookmarked and re-run it the night before every design interview; onboarding a teammate onto a gnarly service, I'd rather hand them a five-question quiz than a forty-minute recording.

Why this actually works

The workflow is quietly doing two jobs at once. First, it flips passive watching into active recall — the exact behavior Freeman's data rewards, and the reason a quizzed hour beats a rewatched one. Second, it hands you the pace control a video never gives you: linger ten minutes on the one concept that won't click and skip the three you already own. That second lever is where the real compounding happens — a 30-minute talk becomes 30 minutes aimed only at what you don't know yet, every time you run it.

Do this today

Pick the talk you keep bookmarking and never actually opening (you know the one). Run all four steps this afternoon, then study it in two 25-minute sessions this week — quizzing yourself, not rewatching. Walk into your next interview able to use the material, not just recall that you once saw it. You'll know it better than 95% of the people who watched the same video and moved on.

Sources