⚠️YouTube Processor Notice:

The YouTube Processor currently only works on localhost due to YouTube API rate limiting (429 errors).

Please run this locally and see the README for setup instructions.

LLM Review Analysis Dashboard

Local Only

YouTube Video Processor

YouTube Video ID or URL

Stream progress

General analysis ("What is this video about?")

Demo LLM Analysis Buttons:Instantly preview the full LLM bias analysis pipeline without hitting YouTube or OpenAI.
These buttons load a real review, transcript, and metadata as if it was processed live, so you can see the exact output and UI for a positive (Mario Wonder) and negative (Concord) review.
No API calls, no rate limits, no YouTube required.

YouTube Batch Processor

YouTube Video IDs (comma, space, or newline separated) or URLs

Mode:LLM Model:

How to use this dashboard

Bar graph: Shows mismatches between AI analysis and available ground truth data. High bars for sentimentScore/verdict are EXPECTED - these are AI-only fields with no baseline data to compare against.
Real Issues to Focus On: Look for mismatches in pros, cons, reviewSummary, alsoRecommends - these fields sometimes have explicit mentions in transcripts that we can verify against.
Grouped View: Each row is a reviewId, with all fields shown as columns. Cell colors: Red = mismatch detected, alternating white/yellow = normal.
Advanced QA: Each row is a single field comparison. Use this for detailed analysis of specific mismatches.
This is an internal QA tool to help us fine-tune our AI analysis before building user-facing features.

Field Mismatches Overview

Note: Each bar shows the number of mismatches (similarity < 0.8) for that field. Higher bars mean more disagreement between LLM and original data (worse). Lower is better.

Show only mismatches

Sweep:

Mismatch?

Overridden?

Alt row?

Normal?

Review ID	Actions

Showing 0 of 0 total reviews