YouTube data collection pipeline for research

This case is positioned as a clean CLI and data-pipeline showcase for collecting YouTube data and publishing stable JSON artifacts for downstream analysis. The emphasis is not just on grabbing subtitles, but on a repo with a clear public contract: cleaner README, runbook, architecture notes, curated examples, and predictable exports.

I built a TypeScript CLI on Node.js with an explicit flow for single-video fetches, batch processing, and channel-level exports. The repository now documents the operational layer as well as the code: a cleaner README/runbook, `docs/architecture.md`, CI, typed response boundaries, and small representative fixtures under `examples/`.

In practice the pipeline had to deal with unstable player responses, platform limits, and long-running batches. The current repo captures that reliability work with retries, backoff, cooldown windows, key and user-agent rotation, deterministic exports, and transparent notes about the current `youtubei/v1/player` adapter, while keeping that endpoint detail out of the headline positioning.

The result is a polished public repo that makes the project contract easy to review: per-video JSON under `video_data/`, combined channel exports under `exports/`, a runbook for `fetch` / `batch` / `export`, architecture notes, and curated examples for quick inspection. It works as both an engineering showcase and a practical base for text-analysis workflows.