Objective:
Automate the evaluation of image and caption submissions to ensure quality, relevance, and compliance with safety standards.
Key Features:
- AI-powered caption analysis using Nvidia’s NeMo Guardrail to filter inappropriate, explicit, or nonsensical content.
- Image assessment using Langsmith’s Image Analyzer to detect NSFW content and assign a quality score.
- Automated text-image relevancy check to verify if captions accurately describe corresponding images.
- A structured scoring mechanism integrates text and image evaluation for comprehensive content moderation.
- Reduction of manual moderation effort while enhancing data integrity and compliance.
Results:
- Enhanced content quality with automated rejection of inappropriate or irrelevant submissions.
- Reliable validation of image-text correlation improves overall data consistency.
- Significant reduction in manual intervention required for content review.
Customization for IT Staff:
- Adaptable for IT service management (ITSM) platforms to analyze documentation, screenshots, and support tickets.
- It can be modified to assess internal documentation quality and accuracy, ensuring technical correctness.
- AI-driven automation to categorize, validate, and improve the quality of IT knowledge base articles.
Timeline 13 Weeks:
- Caption Quality Check (Nvidia’s NeMo Guardrail): 2 weeks
- Image Analysis (Langsmith’s Image Analyzer): 3 weeks
- Text Analysis: 1 week
- Score Calculation: 3 weeks
- Image-Text Relevancy Check: 4 weeks
Techstack:
Python, Flask, Langchain, LLM, GPT-4o, Bing Search, ElevenLabs