Agentic Programming · Lab 6: Make an Explainer Video
A 66-second narrated explainer on binary search, made entirely in code with Claude Code. The animation is Remotion (React → MP4), the voice is local Kokoro text-to-speech, and every visual is drawn with divs and SVG. There are no generated images.
media/binary-search-explainer.mp4 (720p, 1.4 MB)
Takeaway: Binary search finds an item in a sorted list by checking the middle and throwing away half the list each step, so it needs only about log₂(n) checks: 20 checks for a million items.
| # | Scene | What you see |
|---|---|---|
| 0 | Title | "BINARY SEARCH" with a row of cells whose middle one lights up |
| 1 | Hook | A 1,000-item list scanned one by one, with the counter racing to 1,000 |
| 2 | Sorted is the key | 16 scrambled cells sort themselves. Target: 79 |
| 3 | First check | low / mid / high pointers. 42 < 79, so the left half is thrown away |
| 4 | Repeat | 73 → 85 → 79 found: 4 checks instead of up to 16 |
| 5 | Why it's fast | 16 → 8 → 4 → 2 → 1, a chart of n vs log₂ n, then 1,000,000 → 20 |
| 6 | Recap | Python code with four step cards that light up the matching lines. O(log n) |
The full script, with narration and timings, is in SCRIPT.md.
- Narration first.
narration.jsonholds the text for each scene.tts/narrate.pyturns it intopublic/audio/<scene>.wavand writessrc/durations.json, so every scene's length follows its audio. - Animation cues sync to the speech. Pauses in each clip were measured with
ffmpeg silencedetect, and each scene'sCUEtable times its animations to those moments. - One scene per file. Each scene is its own file, so two agent sessions could build scenes in parallel without conflicts.
src/Main.tsxstrings them together with<Series>and<Audio>. - Shared pieces. The colors and fonts are in
src/theme.ts, the example array is insrc/data.ts, and the reusable array row with its pointers is insrc/components/ArrayRow.tsx.
Requires Node 18+, Python 3.10+ and ffmpeg.
npm install
py -3.11 -m venv .venv # or: python3 -m venv .venv
.venv/Scripts/pip install kokoro-onnx soundfile
# download kokoro-v1.0.onnx and voices-v1.0.bin into models/
# (https://github.com/thewh1teagle/kokoro-onnx/releases)
npm run narrate # regenerate the narration (optional; the .wav files are committed)
npm run studio # live preview at http://localhost:3000
npm run render # → out/video.mp4The video was built in small steps with a commit after each one: setup and a 5-second test clip, then one scene per commit, each previewed in Remotion Studio and checked against the audio before moving on. Two Claude Code sessions worked side by side: one wrote the script, the narration and Scenes 1–2, and the other built the shared row component and Scenes 3–6. git log shows the whole sequence.
Part of my Agentic Programming labs (HUJI, AICODE 67723): Lab 1 · CTF Arena · Lab 2 · LLM failure modes · Lab 3 · Coin Dash with several agents · Lab 4 · Traffic monitor spec · Lab 5 · Chess: testing new features · Lab 6 · Binary search explainer video
