How PYLER Uses SingleStore to Power Real-Time Brand Safety and Contextual Intelligence at Scale

PYLER delivers real-time video understanding that helps brands ensure brand safety, contextual relevance, and trust. Its technology interprets video, audio, and metadata to determine whether ads appear in the right context. The company supports major brands through its three flagship products: AiD for end-to-end brand safety validation, AiM for contextual integrity, and Measure for video based market intelligence. Clients such as Samsung Electronics, Kenvue, L’Oréal, and LVMH rely on these platforms to measure, verify, and understand the environments in which their ads appear.

Challenges / Goals

Once brands start using PYLER’s technology, they stay. The scale of data PYLER processes has been growing exponentially as leading brands continue to rely on its solutions. To date, PYLER has analyzed over 700 million videos and now processes more than 2 million videos every day — a number that continues to climb. As the volume of video data has surged, efficient large-scale data management has become mission-critical. To stay ahead, PYLER made a strategic decision to operate at the cutting edge by becoming one of the first startups to deploy NVIDIA B200 GPUs and partnering with SingleStore, a disruptive leader in modern data infrastructure, to ensure performance and scalability at every level.

Before adopting SingleStore, PYLER ran its operations on a combination of PostgreSQL Aurora and AWS Lambda functions. This architecture worked for smaller workloads but struggled as the company expanded. Their primary data sources, real-time ad campaign performance metrics and contextual video analytics, produced an overwhelming stream of structured and unstructured data.

The growing scale exposed serious pain points. “Business-wise, slow reporting and insights made it difficult to deliver real-time campaign feedback to customers,” said Dongchan Park, VP of Engineering. “Technically, Aurora with Lambda required frequent ETL-style batch jobs, which introduced latency and complexity.” Performing joins between live aggregated data and historical datasets became increasingly inefficient and expensive.

Maintaining this pipeline demanded constant optimization. Every dashboard refresh, every query, and every customer report came with the risk of latency spikes. These delays affected PYLER’s customers directly. Brand managers waiting for campaign verification data often faced delays, which meant missed opportunities to adjust placements or protect brand reputation in real-time.

Technology Requirements

PYLER outlined several technical priorities for its new system:

• Real-time ingestion and aggregation across streaming video analytics and ad performance data
• Efficient joins between live and historical datasets for contextual validation
• Unified support for both transactional and analytical workloads in one database
• Query performance under two seconds even for complex, join-heavy analytics
• Scalable and reliable architecture that supports tens of thousands of queries per minute and hundreds of users concurrently
• Operational simplicity so engineers could focus on new features instead of performance tuning

Why SingleStore

The decision to move to SingleStore came after years of tuning PostgreSQL Aurora to its limits. Hyeongjun Park said, “In PostgreSQL, optimizing queries with multiple large table joins and aggregations was very difficult, even with materialized views and partitioning.”

By migrating to SingleStore, PYLER gained the ability to ingest, analyze, and serve data simultaneously. “With SingleStore, careful shard and sort key design delivers excellent query performance along with outstanding ingestion speed for large-scale advertising data,” says Dongchan Park. “This has reduced engineering overhead and operational stress while enabling our team to focus on higher-value initiatives.”

SingleStore’s hybrid transactional-analytical (HTAP/HDAP) architecture proved to be the defining factor. “The most important differentiator for us is SingleStore’s hybrid capability,” says Hyeongjun Park. “It is very rare to find a database that can effectively handle both transactional and analytical workloads at scale.”

The results were immediate and measurable: more than a 10× improvement in data ingestion speed and over a 100× reduction in query latency. These gains directly impacted customers by enabling true real-time validation and campaign measurement.

“Our backend engineers no longer need to spend excessive time and cost on database performance tuning,” notes Dongchan Park. “Instead, they can now focus on delivering new features and strategic value.”

Finally, SingleStore’s architectural flexibility allows PYLER to scale efficiently while preparing for the future. Hyeongjun Park adds, “We currently use pgvector for vector workloads, but we see potential to move to SingleStore’s native vector capabilities as our requirements for larger-scale throughput and lower latency increase.”

Solution

PYLER deployed SingleStore to simplify its data operations and support real-time analytics at scale. Using SingleStore Workspaces, the team can create and manage environments directly through the Helios portal, which they describe as intuitive and easy to deploy. This approach allows PYLER to scale its infrastructure efficiently and manage workloads across services without operational friction.

The company also uses SingleStore Pipelines for S3 backup, ensuring reliable data protection and simplified recovery as their datasets continue to grow. Together, these capabilities give PYLER a streamlined and resilient data architecture that supports continuous performance and operational confidence.

Finally, SingleStore’s separation of compute and storage gives PYLER the flexibility to scale dynamically as workloads fluctuate. During large campaigns, compute clusters expand automatically to meet demand, then contract when volume subsides. This architecture keeps query latency consistently low while maintaining predictable infrastructure costs.

Together, these capabilities give PYLER the performance, stability, and simplicity it needs to deliver real-time brand-safety analytics to some of the world’s most recognized brands.

Outcome

Real-Time Speed at Scale

PYLER now processes millions of rows per second while maintaining performance and stability. Real-time ingestion allows the company to deliver up to the minute brand safety and contextual validation results, helping brands act immediately on data driven insights.

Sub Two Second Query Latency

Complex analytical queries involving joins and group operations now return in less than two seconds, providing interactive dashboards and live feedback for brands.

Ten Times Faster Ingestion and Hundred Times Faster Queries

PYLER achieved more than a tenfold improvement in data ingestion and a hundredfold reduction in query latency, allowing real-time analytics at petabyte scale.

Greater Engineering Efficiency

SingleStore simplified data operations and freed engineers to focus on product innovation rather than tuning.

Flexible Scaling and Cost Control

With separation of compute and storage, PYLER scales elastically during peak campaign loads and optimizes infrastructure costs when usage declines, ensuring predictable performance and spend.

Summary

By adopting SingleStore, PYLER unified its transactional and analytical workloads, eliminating batch latency and enabling real-time analytics at global scale. The company now delivers sub two second queries, faster ingestion, and simplified operations. With SingleStore as its foundation, PYLER continues to expand AI driven video understanding and brand safety analytics for some of the world’s most recognizable brands.