Accelerating Video Understanding for ad generation with Visual AI Agents

PYLER supercharges video ad safety with NIVDIA's DGX B200 and AI Blueprint for smarter, safer placements.
Discover how PYLER's real-time video understanding helps brands like Samsung, BYD, and Hana Financial thrive.

At PYLER, we leverage our platform, video understanding AI, which deploys visual AI agents to ensure brand safety by verifying that ads are never displayed alongside inappropriate videos and placing them in more contextually relevant settings to protect brand value and enhance campaign effectiveness.

With the explosive growth of generative AI, the sheer volume of content uploaded to video platforms has skyrocketed. In such an environment, displaying ads next to content that’s not brand aligned increases marketing costs to target the correct audiences, but also presents a risk of negative brand impressions and potential reputational damage.

Video Understanding AI is crucial for effectively managing and validating content in an era of massive, high-speed uploads. When properly implemented, it becomes a key asset in safeguarding and enhancing a brand’s overall value.

Video content processing is significantly more complex than image or text processing, demanding substantial computational resources from preprocessing to inference. To handle this real-time influx of video content effectively, a high-performance pipeline utilizing optimized computing resources is essential.

In this post, we’ll discuss how PYLER addresses these challenges by adopting NVIDIA AI Blueprint for video search and summarization (VSS) in conjunction with NVIDIA DGX B200 systems. We’ll explore how we pair NVIDIA software and accelerated computing, with PYLER’s proprietary Video Understanding AI, to help ensure brand safety at scale.

PYLER Boosts Content Safety with NVIDIA AI Blueprint for VSS and NVIDIA AI Enterprise

PYLER automatically filters, classifies, and analyzes large-scale video data to help advertisers run their campaigns safely and efficiently. To accomplish this, PYLER selected NVIDIA AI Blueprint for VSS, built on the NVIDIA AI Enterprise software platform, offering enterprise-grade support, advanced security measures, and continuous, reliable updates.

NVIDIA AI Enterprise delivers innovative blueprints and commercial stability, enabling PYLER to deploy secure, scalable and continuously maintain AI workflows.

Key advantages provided by VSS Blueprint include:

1. High-Quality Video Embeddings

   • State-of-the-art embedding models are essential for capturing rich contextual and semantic information from videos. PYLER integrates the NVIDIA NeMo Retriever  nv-embed-qa-1b-v2 model for minimal information loss during the embedding and captioning process.

2. Scalable Infrastructure for training custom VLMs and Large-Scale Video Processing

   • NVIDIA DGX B200 accelerates custom VLM training and delivers up to 30x faster performance and best-in-class energy efficiency compared to its predecessor. 

   • With the immense volume of video content uploaded daily, GPU-based parallel processing is critical for scalability. PYLER utilizes VSS blueprint’s chunk-based parallel processing and GPU node expansion to dynamically distribute workloads and process video data in real-time. We look forward to enabling this on NVIDIA DGX B200 for inference.

3. Flexible Verification and Search via RAG

   • PYLER rapidly searches and verifies vast amounts of video embeddings, leveraging Context-Aware Retrieval-Augmented Generation (RAG). This supports automated summarization, indexing, Q&A, and content validation tasks, significantly improving inference speed, accuracy, and context-awareness.

     
     

Expected Outcomes for PYLER

With these integrations, PYLER’s Video Understanding AI software can provide:

1. Maximized Brand Safety

   • Precisely filter, classify, and recommend safe ad placements, ensuring ads never appear next to inappropriate or harmful content.

2. Enhanced Brand Trust

   • Provide advertisers with robust and reliable AI models, fostering stronger viewer trust. As brands gain consumer confidence, PYLER’s credibility as an industry leader grows in parallel.

3. Improved Data Pipeline Efficiency

   • Deploy automated, GPU-accelerated workflows optimized at a scale far beyond traditional MLOps pipelines, significantly increasing operational efficiency and throughput.

4. Faster Service Launch

   • Leverage NVIDIA’s cutting-edge technology stack – including VSS blueprint, NVIDIA Metropolis, NVIDIA Dynamo-Triton, formerly Triton Inference Server, , NVIDIA NeMo Retriever and DGX B200 systems, – enabling innovative enterprises to quickly launch and scale AI-powered video services with fewer resource constraints.

     
     

Customer Success Stories

1. Samsung Electronics: PYLER AiD Redefines Brand Safety Standards for Product Branding Campaigns

• As a leading global brand, Samsung Electronics faced challenges protecting its brand from appearing alongside misaligned content on digital platforms like YouTube
  By integrating PYLER AiD’s real-time AI video analysis, Samsung Electronics automatically blocked ads from harmful content categories based on global standards.
  By integrating the AiD solution into Samsung Electronics' ad campaigns, ads were automatically prevented from running on content categories threatening brand values, such as violence, explicit content, hate speech, and fake news. The AiD dashboard ensured brand safety and trust with transparent real-time monitoring and performance tracking.

• Results:

  • Significant Reduction in Harmful Content Exposure: PYLER AiD reduced ad exposure to sensitive or inappropriate content that could threaten brand value by 77%, setting a new standard for brand protection.

  • Optimized Ad Budget Spend: Preventing ads from appearing on unsuitable content allowed Samsung to focus its marketing budget on safe, high-value, brand aligned placements, greatly increasing advertising effectiveness.

  • Enhanced Brand Trust and Value: Proactively eliminating potential brand crises and consistently delivering a safe and trustworthy brand image to consumers successfully defended and strengthened brand reputation and customer trust.

2. BYD: PYLER is Ensuring Successful Korean Market Entry Through Brand Protection and High Intent Moment Targeting.

• As a global leader in electric vehicles, BYD faced the critical task of successfully launching its first model in Korea, the 'Atto 3'. It was essential to create a positive first impression while proactively managing the risk of ads appearing alongside potentially negative content (e.g., critical reviews, defect discussions) that could harm the new brand's image upon market entry.

• PYLER's AI-Powered Contextual Targeting and Brand Suitability Management technology precisely targeted BYD ads for positive or neutral YouTube content about EVs and small SUVs such as vehicle reviews, test drives, and competitor comparisons. Simultaneously, the AI excluded negative videos by analyzing content tone and context. This ensured BYD ads appeared only in positive, relevant high intent moments aligned with potential customer interest, effectively achieving both contextual targeting and brand suitability simultaneously.

• Results:

  • Maximized Exposure in Brand-Friendly Contexts: Ad placements within safe, positive, and relevant EV/SUV video contexts increased exponentially—more than 100x— compared to previous methods—delivering high relevance and brand suitability to potential customers.

  • Significantly Boosted Potential Customer Interest: By effectively filtering out negative noise and concentrating ads on highly relevant content, BYD captured genuine potential customer interest, driving click-through rates (CTR) nearly 4x higher than baseline campaigns.

  • Successful Market Entry & Enhanced Brand Image: PYLER's integrated AI solution effectively managed the risks associated with new market entry, built a positive brand image, and played a key role in BYD's successful initial launch in the Korean market, receiving over 1,000 pre-order applications in the first week.

3. Hana Financial Group: PYLER Moment Optimizes Performance Across Diverse Campaigns with Automatic Video Placement Recommendation and Targeting

Hana Financial Group, one of South Korea’s largest financial holding companies manages a diverse portfolio including banking, credit cards, securities, and asset management. Facing the challenge of running multiple YouTube campaigns with distinct goals-such as attracting public pension accounts, providing retirement pension information, and promoting a public service announcement against illegal gambling-the group needed to reach targeted audiences within relevant content contexts while achieving varied KPIs like conversion rates, ad efficiency, and user engagement.

To address this, Hana Financial Group implemented PYLER Moment, an AI-powered contextual targeting solution. PYLER’s AI video analysis enabled precise targeting of specific YouTube content moments aligned with each campaign’s objective. For example, the Public Pension campaign focused on retirement planning and celebrity ambassador content, while the Retirement Pension campaign targeted videos about post-retirement finance. The Anti-Gambling PSA featured esports star Faker and targeted gaming and esports content popular with younger viewers, maximizing engagement and campaign effectiveness.

• Results:

  • Optimized Performance for Each Campaign Goal:

    • (Public Pension) Achieved high average conversion rates exceeding 1.5%, demonstrating effective customer acquisition, with specific target groups surpassing an outstanding 3.6% conversion rate.

    • (Retirement Pension) Showcased dramatic improvements in ad efficiency compared to general targeting, with click-through rates (CTR) increasing more than eightfold and cost per click (CPC) reduced by over 70%.

    • (Public Service Ad) Successfully connected the featured celebrity 'Faker' with fan interests, resulting not only in a noticeable uplift in CTR but also proving high user immersion, with average view duration per impression and 100% video completion rates doubling compared to general campaigns. Furthermore, it reached more unique users efficiently, with average frequency halved.

  • Accurate Target Channel Reach: Across all campaigns, PYLER's contextual targeting concentrated ads within highly relevant channels, demonstrating significantly better reach within key target channels compared to general targeting. This confirmed the effective allocation of the ad budget to the most suitable potential customers.

    
    

Conclusion and Future Outlook

By adopting agentic AI through NVIDIA AI Blueprint for video search and summarization and DGX B200 systems, PYLER aims to strike a balance between protecting brand value and maximizing advertising efficiency in a market flooded with video content.

With large-scale GPU parallel processing, efficient embeddings, and Context-Aware RAG for search and verification, we will be able to detect negative content almost in real time and reliably adhere to brand guidelines.

Going forward, PYLER plans to continue actively harnessing NVIDIA’s latest AI technology to deliver fast, high-quality services that connect video content safely and appropriately across various domains.