🛡ī¸

Trust & Safety Analyzer

Internal & confidential — enter the access password.

🛡ī¸ Trust & Safety Policy Analyzer

INTERNAL Confidential — Do not share results externally
📋 Analyzer
⚖ī¸ Compare
📝 Policies
đŸ’Ŧ Chat
🎙ī¸ Audio
đŸ–ŧī¸ Images
đŸ”Ŧ Workbench
📊 History
📚 Decision Library
📈 Metrics
🔧 Prompts
ℹī¸ Info

Analyze Content

Paste text content and evaluate it against policies you've uploaded in the Policy Editor. Select content type, choose policies, and optionally toggle decision mode.

General Lyrics Transcript Podcast Book Excerpt Conversation Blog/Article
No custom policies — general T&S policy will be used
Off: Evaluation only — observations, flags, and concerns for you to decide.
On: Decision mode — AI makes the call on whether to act, with a recommended action.

Content Analyzer

Paste or import content on the left and click Analyze.
The AI will assess it against your loaded policies for violations, sentiment, tone, sarcasm, and intent.

đŸ“Ļ Bulk Analyzer

Analyze multiple content items at once. Upload a CSV/text file or paste items (one per line). Results exported as a table.

Bulk Analyzer

Paste or upload content items on the left. Each line = one item analyzed separately.
Results appear as a table you can export.

⚖ī¸ Side-by-Side Compare

Analyze the same content with two different models simultaneously to compare their assessments.

Model Comparison

Enter content and select two models to see how they assess the same content differently.

Your Policies

🔒 Private to you. Policies are used by the Analyzer to evaluate content.

Policy Editor

Click + Create New Policy or select one from the list to edit it.

You can type, paste text, or drop/upload images (PNG/JPG) and the text will be extracted automatically.

Chat Settings

Discuss policy interpretations, edge cases, and get assessments from the AI assistant.

No policies loaded
Hi! I'm your Trust & Safety policy assistant. I can help you understand policy points, discuss edge cases, and make assessments. What would you like to discuss?

🎙ī¸ Audio Content Extraction

Upload audio files (podcasts, songs, conversations) to transcribe them. Once transcribed, send the text to the Content Analyzer for policy review.

đŸŽĩ

Drop audio file here or click to upload

Supports MP3, WAV, M4A, OGG

⚠ī¸ Please use with leadership approval — audio transcription incurs costs (~$0.024/min).

🎙ī¸ Amazon Transcribe — Active
Upload audio and it will be automatically transcribed.
â€ĸ Supports MP3, WAV, M4A, OGG, FLAC
â€ĸ Speaker identification enabled (up to 10 speakers)
â€ĸ Transcription typically takes 1-3 minutes
â€ĸ Once complete, transcript appears below — send it to the Analyzer for policy review

Manual Transcript Entry

If you have a transcript from a podcast or song, paste it here and send it to the analyzer.

đŸ–ŧī¸ Image Analysis

Upload or paste images to evaluate against your policies. Supports multiple images. No guardrails — designed for reviewing potentially offensive visual content.

đŸ–ŧī¸

Drop images here or click to browse

PNG, JPG, GIF, WEBP — multiple files supported

📋 Click here and Ctrl+V to paste images

Image Analysis

Upload, drop, or paste images on the left to analyze them against your policies.
Images are analyzed directly by AI vision — no guardrails applied.

đŸ”Ŧ Workbench

Batch review and label content for ML training, or generate images. Import content, label items, and export datasets.
📋 Review & Label
🎨 Generate Image
Items: 0 | Labeled: 0

Workbench

Import content to review and label, or generate images.
Use Review & Label for batch content work, or Generate Image for image creation.

Analysis History

⏱ī¸ Your analyses only — stored for 30 days then auto-deleted.

Analysis History

Select an analysis from the left to view its full results.
Results are private and not shared. Use Export to save.

📚 Decision Library

Shared across all users. Feedback is stored here and used to inform future evaluations. Feedback only applies when the same model and same policy selection are used — ensuring decisions are consistent within each model/policy combination.

ℹī¸ Info & Costs

💰 Cost Breakdown

Service Cost Notes
Bedrock (Claude Sonnet 4) ~$0.02 per analysis Only when Analyze or Chat is used
Amazon Transcribe ~$0.024 per minute of audio ~$1.44 for a 60-min podcast
Bedrock — Image Analysis ~$0.05–$0.10 per image Images use ~1,000–1,600 tokens each depending on size
Bedrock — Image Generation ~$0.04 per image (Nova Canvas) / ~$0.01 per image (Titan) Workbench image generation, per image generated
DynamoDB ~$0.00 (pennies) Pay-per-request, negligible at this scale
Lambda ~$0.00 (free tier) 1M free requests/month
API Gateway $1 per million requests Essentially free at normal usage

📊 Estimated Monthly Cost

Light usage (10 analyses/day): ~$6/month
Moderate usage (50 analyses/day): ~$30/month
Heavy usage (200 analyses/day): ~$120/month
Audio transcription: add ~$1.44 per hour of audio transcribed

Zero cost when nobody is using the tool.

🛡ī¸ About This Tool

Trust & Safety Policy Analyzer — AI-powered content moderation assistant

Features:
â€ĸ Analyze content against uploaded policies (lyrics, transcripts, podcasts, books, conversations, blogs)
â€ĸ Sentiment analysis, intent detection, tone analysis, sarcasm detection
â€ĸ Policy chat assistant for discussing edge cases and interpretations
â€ĸ Audio transcription via Amazon Transcribe (auto speech-to-text)
â€ĸ Configurable Bedrock guardrails for handling explicit content
â€ĸ Persistent policy and history storage
â€ĸ Export analysis results as JSON

Powered by: Amazon Bedrock (Claude Sonnet 4), Amazon Transcribe, DynamoDB, Lambda

⚠ī¸ Policies are confidential and internal. Do not share analysis results externally.

⚙ī¸ Guardrails

✅ All guardrails are disabled across all features. You have unrestricted access for Trust & Safety content moderation work. The Prompt Attack filter shows LOW due to an AWS minimum requirement but is not enforced on any model calls.

🔧 Prompt Editor

Edit the system prompts that control AI behavior across the app. Changes take effect immediately. Password protected.

This section is password protected.

📈 Metrics Dashboard

Overview of analysis activity, violation rates, and model usage. Data from history (last 30 days).