Safari AI Cuts Cloud Costs by 50% While Scaling Real-Time Computer Vision Analytics with StreamNative

Featuring: Kaiwen Yuan, Co-Founder, Head of Engineering at Safari AI

"StreamNative's resilience is critical to our SaaS operations. Each quarter, we evaluate Pulsar, StreamNative, and other solutions, and StreamNative consistently emerges as the best choice—cutting our costs by more than 50% while seamlessly supporting our ML data structure requirements. We're also collaborating closely with the StreamNative team to adopt innovations like Ursa Engine, Iceberg integration, and more to sustain our growth and profitability" - Kaiwen Yuan, Co-Founder, Head of Engineering at Safari AI

Background

Safari AI helps enterprises understand their physical operations by digitizing real-world activities. Using a customer’s existing camera infrastructure, the company provides automated measurements of critical physical activities via Computer Vision technology, helping businesses track metrics like guest occupancy, staff engagement, parking utilization, and queue wait times. This real-time operational intelligence enables managers and front-line staff to focus on customer service rather than manual monitoring tasks.

Challenge

Safari AI faced several critical challenges in building their computer vision platform:

• Processing massive amounts of video data efficiently while maintaining the company’s commitment to 90-95% accuracy.
• Managing costs while scaling to handle 10,000+ pipelines and 50,000+ cameras.
• Ensuring real-time data delivery without requiring video reprocessing when camera positions change.
• Finding a cost-effective way to store and process ML data without maintaining expensive infrastructure.
• Previous solutions like Kafka and AWS Kinesis required significant DevOps resources and were costly to maintain.

Solution

‍ Safari AI implemented StreamNative's fully managed platform as their primary data storage and streaming backbone, creating an end-to-end solution for real-time operational intelligence. StreamNative’s comprehensive platform collects ML data from edge devices and processes this information through Safari's Flink-based data engine, enabling immediate delivery of metrics and alerts to customers. By leveraging StreamNative's tiered storage capabilities, Safari AI achieved cost-effective data retention while maintaining rapid access to historical information. The platform's utilization of StreamNative's schema registry proved particularly valuable for handling structured ML data, ensuring consistent data quality and format across their growing deployment of computer vision analytics. This integrated approach enabled Safari AI to focus on delivering value to their customers without the overhead of managing complex streaming infrastructure.

‍Technical Journey

‍Safari AI's architecture consists of four main components:

1. Customer's existing camera infrastructure;
2. GPU server for ML services;
3. StreamNative Cloud for data streaming and storage; and
4. Safari AI Cloud for business metrics processing.

The platform processes video feeds through ML models, sends the output with timestamps and IDs to StreamNative's Pulsar, and then transforms this data into actionable business metrics through their data engine.

‍Results

‍The implementation of StreamNative's platform delivered significant operational and financial benefits for Safari AI. Most notably, the company achieved a 50% reduction in infrastructure cost compared to their previous solutions, while maintaining consistent sub-10 second end-to-end latency for real-time metrics delivery. The switch to StreamNative's fully managed service eliminated the need for a dedicated DevOps team to manage streaming infrastructure, further reducing operational overhead. The platform's efficient storage capabilities enabled Safari AI to maintain one year of back-processing storage, providing valuable historical data access for their clients. Most importantly, the solution successfully scaled to support multiple enterprise clients through shared resources, demonstrating the platform's ability to grow with Safari AI's business needs while maintaining cost efficiency. These results validate Safari AI's decision to choose StreamNative as their core streaming infrastructure provider, enabling them to focus on their core mission of delivering computer vision analytics to their customers.

‍Key Takeaways

• Safari AI archived a 50% cloud cost reduction compared to their previous solutions, while maintaining extreme low latency < 10ms and longer data retention.
• Schema compatibility is crucial for ML data structure requirements.
• The multi-tenancy hierarchy supports various clients while sharing underlying resources.

Future Prospects

Safari AI plans to adopt StreamNative's Ursa Engine and Iceberg integration, which, based on its initial evaluation and investigation, shows promise in further reducing costs—by up to 90%—for new client implementations while sustaining growth and profitability. The company will continue expanding its no-code platform and auto-calibration capabilities to enhance the accessibility and scalability of computer vision analytics.

‍Conclusion

By choosing StreamNative over other alternatives like Confluent and Redpanda as their streaming platform, Safari AI has built a scalable, cost-effective solution for computer vision analytics that delivers consistent accuracy and performance for their enterprise customers. The partnership enables Safari AI to focus on their core mission of digitizing physical operations while maintaining reliable, real-time data delivery.