The visible feature was only the surface
Viewers must scrub long YouTube videos to find specific answers even when the page already exposes metadata and a transcript.
The system also had to respect concrete constraints. Answers are limited by transcript availability and quality. Gemini calls require the user's own API key and send bounded video context to an external provider. YouTube route/DOM changes and extension permission boundaries can break extraction.
My role covered product design and browser-extension engineering, with responsibility for browser integration, provider workflows, and privacy.
I turned the constraints into boundaries
YouTube content extraction; React extension UI and request state; validated transcript/context envelope and Gemini adapter; browser storage/permissions plus release-built Manifest V3 package. The companion web app documents and distributes the product.
Use visible page metadata/transcripts rather than downloading or processing media.
Keep API credentials user-supplied instead of operating a shared proxy, trading onboarding friction for a smaller service/data boundary.
Separate extension and website workspaces so product runtime and public documentation can ship independently.
The key decision: Use visible page metadata/transcripts.
The implementation had to prove the model
I delivered the work across the full path: Built YouTube route detection and transcript/metadata extraction. Implemented Gemini-backed question answering with schema validation and bounded context. Added local credential/settings handling and explicit privacy guidance. Created release automation and extension/web quality gates.
The result is concrete: Users can ask questions about an open YouTube video. Answers can use available transcript and visible metadata. Users control their own provider credential. Maintainers can produce installable GitHub release artifacts.
The senior engineering lesson was that solving the visible workflow is only half the job. The architecture must also make constraints, failure modes, evidence, and ownership explicit enough for the next change to remain safe.