How Loka uses Amazon Nova Pro to improve video surveillance and event detection
Challenge
A major enterprise monitoring constant streams of video was overwhelmed by the massive volume of visual events it received each day. Millions of non-relevant moments had to be manually reviewed by a large team of human operators. Without an intelligent pre-filtering system, the process was slow, expensive, and prone to delays in identifying critical footage. The company needed to reduce operator load while improving threat detection.
Solution
Loka used Amazon Nova Pro to build a powerful AI tool that automatically identifies and classifies millions of visual events in video footage. Amazon Nova Pro provides the best combination of accuracy, speed, and cost. Built on Amazon Bedrock, Loka developed a tool that accurately distinguishes routine or benign events, effectively filtering out non-critical activity and alerts. This enables human operators to focus on high-priority footage, helping them detect threats faster and more efficiently. The use of generative AI enhances customer safety and satisfaction while significantly reducing human workload.


Built using Amazon Nova
Amazon Nova Pro reduced irrelevant alerts by 55% while maintaining a threat detection rate above 97%. By automating first-pass filtering, Loka enabled its client to operate with a level of precision and speed that far surpassed manual review, doubling video monitoring efficiency. Not only did the solution outperform manual processes, it also delivered strong overall performance compared to other models. Nova Pro was ultimately selected for providing the right balance of accuracy, efficiency, and cost-effectiveness for the customer’s needs.
Amazon Nova
As an AWS Premier Partner and 2024 Innovation Partner of the Year, Loka builds generative AI and cloud-native solutions that power startups and enterprises across industries. With a track record of moving from POC to production at one of the highest rates in the AWS ecosystem, Loka turns prototypes into long-term value.