Digital Asset Management Metadata Tagging solution using AI

Overview: 

As modern enterprises scale, the volume of digital assets (marketing, brand, product) in their Digital Asset Management (DAM) systems explodes. Manual metadata entry, essential for asset discoverability, becomes slow, inconsistent, and resource-intensive, halting scalability.

Acheron developed an AI-powered metadata tagging solution to automatically generate and enrich metadata for images and videos. The solution ensures superior asset findability, compliance, and search accuracy, all while maintaining full data control within the customer’s secure infrastructure.

Challenges

The client faced significant bottlenecks due to the growing complexity of metadata management:

  • High Operational Cost: Manual tagging required extensive human effort for large volumes of assets coming from various internal teams and external agencies.

  • Metadata Inconsistency: Quality varied significantly across teams, leading to inaccurate search results and poor taxonomy alignment across the organization.

  • Complex Tagging Requirements: Metadata models evolved from simple configurations to hundreds of fields per asset type, with 15–20 fields often mandatory for every upload.

  • Security Concerns: Most built-in DAM AI features relied on generic cloud-based models, which posed significant data privacy and security risks for proprietary enterprise assets.

Objectives

The project aimed to transform metadata management from a manual bottleneck into an intelligent, scalable, and secure automated process:

  • Automate Metadata Generation: Achieve up to an 80% reduction in manual tagging effort for images and videos.

  • Ensure Data Privacy: Design and deploy the AI solution entirely on-premise or within the customer’s private cloud infrastructure.

  • Improve Consistency: Enforce high, standardized metadata quality across all asset types for better search and compliance.

  • Enable Human Control: Incorporate a user validation step to ensure accuracy and continuous learning, mitigating potential AI errors.

  • Support Complex Models: Dynamically support multiple, complex metadata models across different asset categories.

The Solution

Acheron built an on-premise, enterprise-grade AI metadata tagging system that analyzes visual and contextual cues, predicts metadata values, and requires user validation before final submission.

Key Features:

  • On-Premise Model Deployment: The entire AI engine is deployed using Docker and Kubernetes within the client’s infrastructure, ensuring zero data exposure to public AI models.

  • AI-Assisted Tagging: Uses advanced Computer Vision models (SSD MobileNet V2, YOLO V8) for highly accurate object detection and classification in images and video frames.

  • Custom Review UI (Human-in-the-Loop): A dedicated user interface allows human reviewers to quickly validate or correct the AI-predicted tags, ensuring accountability and feeding corrections back into the learning loop.

  • Multi-Model Support: The system dynamically maps and tags fields across different metadata models, supporting various asset types seamlessly.

  • Video Analysis: Leverages FFmpeg to extract key frames from video assets, enabling intelligent tagging of video content as well as static images

How It Works

The implementation followed a rigorous 10-step framework focused on performance and security:

    1. Discovery & Field Definition: $\mathbf{22}$ critical metadata fields (mandatory and key searchable fields) were defined for AI automation.

    2. Data Curation: $\mathbf{70\%}$ of historical asset data was meticulously curated, cleaned, and annotated (using tools like RoboFlow) to create high-quality training datasets.

    3. Model Training: The selected Computer Vision models were trained using the curated data, with performance optimized iteratively to meet required accuracy, precision, and recall metrics.

    4. On-Premise Deployment: The finalized models and inference engine were containerized (Docker) and deployed via Kubernetes within the customer’s secure network.

    5. DAM Integration: On asset upload, the system automatically routes the asset to the AI engine.

    6. Prediction and Validation: The AI generates metadata predictions, which are then routed to the Custom Review UI. Humans quickly review and validate the suggested tags.

    7. Finalization: Upon validation, the complete and consistent metadata is pushed to the DAM system, dramatically reducing the asset-to-publish cycle time.

Benefits

  • The solution provided critical strategic benefits related to cost, compliance, and scale:

    • Enterprise Data Security: The on-premise deployment guarantees that sensitive proprietary assets never leave the secure customer environment, meeting strict enterprise security standards.

    • Cost and Time Efficiency: By automating the most time-consuming task in DAM, resources are freed up to focus on content creation and strategy.

    • Scalable Taxonomy: The system dynamically adapts to complex and evolving metadata models, ensuring the DAM platform remains scalable as the business and its content complexity grow.

    • Human-Verified Quality: The Human-in-the-Loop review process ensures the high accuracy of AI predictions while providing a continuous learning mechanism for the models.

Technology Stack

The Outcome

The AI-powered metadata tagging solution delivered measurable improvements across DAM operations:

  • 80% reduction in manual tagging effort, significantly lowering operational costs.

  • Improved search and retrieval accuracy due to consistent, high-quality metadata.

  • Faster asset ingestion and publishing cycles, accelerating time-to-market for marketing campaigns.

  • Improved compliance and taxonomy alignment across vast content repositories

NEWS LETTER

Happenings @Acheron