Holla

Scenario

In the real-time 1v1 video chat scenario, Holla Group faces severe operational and compliance challenges:
1. High-Risk Compliance Requirement: Real-time video streams are highly susceptible to content containing minors, pornography, or other explicit material that severely violates laws (such as COPPA) and platform policies, requiring ultra-real-time detection and enforcement.
2. User Experience and Recommendation: To enhance the precision of in-app user matching, the business recommendation algorithm requires accurate image attribute tagging (e.g., gender, age, image quality) as input.
3. High Concurrency and Low Latency: Video frame auditing must be completed within milliseconds to ensure that non-compliant content can be intercepted instantly (stream cutting) without disrupting the chat experience for legitimate users.

Solutions

DeepCleer constructed an integrated intelligent visual recognition solution for Holla Group, combining "Compliance Safety" with "Business Empowerment":
1. Full-Volume Video Frame AI Auditing: All video frames collected from the client side are sent for auditing. DeepCleer's Intelligent Image Recognition product precisely identifies risks across dimensions including political content, pornography, terrorism, contraband, and facial recognition.
2. Hierarchical Enforcement and Real-Time Stream Cutting:- Images judged as REJECT trigger an instantaneous "real-time stream cutting" action, blocking the violating broadcast and immediately banning the account. This is followed by manual review to determine whether to unban.- Images judged as REVIEW for severe violations (e.g., minors in pornography) are routed for efficient manual review, ensuring no high-risk content is leaked.
3. Business Recommendation Tagging Empowerment: Visual recognition technology is utilized to identify and tag gender, age, and image attributes in normal video scenes. This structured data generated by risk control is returned via API to the client to optimize user matching algorithms.