Fix timeout leak, transcript_segments shape mismatch, and ENABLE_VERTEX_AI boolean guard

[Copilot]

Four bugs identified in code review of the frontend video pipeline PR.

Changes

• Timeout never cleared on fetch error (video/route.ts, transcribe/route.ts): clearTimeout was placed after await fetch(...), so aborts and network errors left the timer pending. Replaced with .finally(() => clearTimeout(timeout)):

  const response = await fetch(url, { signal: controller.signal })
    .finally(() => clearTimeout(timeout));

• transcript_segments always 0 (video/route.ts): Backend returns transcript as { text, source, segments }, not an array. result.transcript?.length was always undefined. Fixed to result.transcript?.segments?.length.

• ENABLE_VERTEX_AI=0 still triggered Vertex AI import (gemini_service.py): Raw os.getenv("ENABLE_VERTEX_AI") is truthy for any non-empty string including "0". Replaced with an explicit allowlist check:

  os.getenv("ENABLE_VERTEX_AI", "").lower() in {"1", "true", "yes"}

• Duplicate let transcript = '' (video/route.ts): Removed duplicate declaration that would cause a compile error.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
v0-uvai	Error		Feb 27, 2026 11:11pm

[railway-app]

🚅 Deployed to the EventRelay-pr-34 environment in EventRelay

Service	Status	Web	Updated (UTC)
supabase	✅ Success (View Logs)		Feb 27, 2026 at 11:05 pm
web	✅ Success (View Logs)	Web	Feb 27, 2026 at 11:05 pm
EventRelay	✅ Success (View Logs)	Web	Feb 27, 2026 at 11:05 pm
guides	✅ Success (View Logs)	Web	Feb 27, 2026 at 11:04 pm
eslint-config	✅ Success (View Logs)	Web	Feb 27, 2026 at 11:04 pm

[vercel]

Additional Suggestion:

Using relative URLs in server-side fetch() calls causes failures in production deployments like Vercel because the URLs cannot be reliably resolved without an absolute origin.

[Copilot]

Pull request overview

This PR aims to address several correctness issues in the video → transcript → analysis pipeline, including timeout cleanup, aligning transcript_segments with the backend transcript shape, and fixing env-var boolean handling for optional Vertex AI imports.

Changes:

• Ensures abort timeouts are always cleared even when fetch() fails/aborts.
• Fixes transcript_segments counting to match the backend’s transcript = { text, source, segments } shape.
• Hardens the ENABLE_VERTEX_AI flag parsing so values like "0" don’t enable Vertex AI behavior.

[Copilot]

origin is declared but never used, and the surrounding comment suggests the code is avoiding deriving an origin from the incoming request. As written, the route still derives baseUrl from request.url for the internal /api/transcribe and /api/extract-events calls, so this is dead code and potentially misleading. Either remove origin/the comment, or actually use a trusted, configured app origin when building those internal URLs (and rename the variable accordingly).

Suggested change

	// Use trusted backend origin instead of deriving from potentially user-controlled request data
	const origin = BACKEND_URL;