Tech:

Phase 1

Models:

-Decoding video to a select number of frames which can be ingested by our CNN model

-Create a CNN model to identify key malicious content (certain % of skin, children, negative gestures)

• train the model with a number of videos/images so it knows what to look for

Model Considerations below

Model Creation

Machine Learning Feedback Loop:

• System for users to provide feedback on false positives/negatives.
• Integration of user feedback to continuously improve the detection model.

API creation

• Allow other systems to connect to the models/tool

Portal

• Results Page

  • A table with all videos/images and their Priority
    • Priority is based on the level of severity which can be set by cleints
  • Filter by priority and also the different malicious content e.g filter swearing/nudity etc

  Legal Compliance and Documentation:

  • Provide legal documentation and terms of use.

  Phase 2:

Dashboard view

• Number of flagged videos in past 30 days / 7 days
• Pie chart of malicious content breakdown
• Total number of videos processed.