Published: Jun 30, 2025

NCS KAI Video Search and Summarisation Platform (KAI SnS)

Video content now comprises nearly half of internet activity, generating vast amounts of video data. Without proper tools, much of this video data remains unused, despite consuming significant space and bandwidth. To utilise information from these large video collections, we need to convert visual information into text for human and algorithmic analysis with accurate temporal and spatial understanding of the videos. AI advancements have created Vision Language Models (VLMs) that can now turn images and videos into text based on input prompts with a coherent understanding of the videos. VLMs enable video summarisation and search, allowing computers to generate concise summaries and locate details in long videos responding to human language queries mapping the temporal and spatial information accurately. This enables various Agentic AI capabilities for applications in various industries including public safety and commerce.
 

In collaboration with NVIDIA, NCS KAI Video Intelligence adopted the NVIDIA AI Blueprint for Video Search and Summarisation from the NVIDIA Metropolis platform to enhance video analytics through AI-powered smart video search. The solution, called KAI SnS, optimises the entire video search and summarisation process using NVIDIA NIM microservices for VLMs, LLMs and RAG. It also supports natural language conversion for efficient data retrieval.


 

Efficient summarisation

• Multimodal indexing employs a knowledge graph with semantic embedding vectors to enable precise, context-aware searches across various videos.
• Video-based generative AI technology utilises cutting-edge vision language model technology to generate dense captions from multiple segments of video, storing them as semantic embedding vectors.

Data representation

• Compact semantic representation uses advanced embedding techniques to represent video content compactly, thereby improving storage efficiency.
• Knowledge graph representation facilitates the identification of real-world entities (people, objects and events) and their relationships. It provides a contextual framework for in-depth understanding by linking semantic video segments to metadata for rapid retrieval. Users can quickly access relevant information even from extensive video datasets.

Smart interactive search

• Uses industry-standard RAG with semantic representation and knowledge graphs to synthesise responses from retrieved content into user-friendly summaries, visualisations and actionable insights.
   

Accelerates development and deployment time

• Enable the deployment of KAI SnS 30% faster with ready-to-deploy Helm charts.
• Flexibly configure and customise NVIDIA SnS workflows with common APIs.
• Enhance accuracy and performance by fine-tuning AI models.

Delivers cost savings

• Reduces manual operation time by a factor of 80x by using AI to efficiently ingest videos, extract and summarise essential information and queries.
• Reduce computing needs by deploying it on a single NVIDIA Hopper GPU and scaling through a highly optimised inference pipeline based on business needs.
   

Public safety

Information summarisation: Turn hours of video from stationary and mobile camera sources into searchable, summarised clips. This makes it easy to find key moments quickly. This helps emergency teams, disaster managers and investigators act faster and make better decisions. For example, you can instantly review the results of an unexpected event without wasting time on manual searches.

Anomaly detection: Actively monitors for unusual or suspicious activities rather than just recording events, sending alerts for abnormal crowd behaviour, unauthorised entries or aggressive actions. Real-time alerts help security teams respond quickly, adding an extra layer of protection.

Commerce

Retail store analysis: Analyses popular product displays, section conversion rates and customer sentiment to improve satisfaction and personalise shopping experiences.

Improve store performance: Understands customer preferences to help retailers make data-driven decisions that enhance sales strategies, customer retention and profitability.

Conclusion

Integrating this powerful KAI SnS (Video Search and Summarisation) tool into enterprise operations unlocks valuable insights, improves decision-making and enhances overall efficiency. Whether for public safety or retail management, it offers a robust solution for analysing and interpreting video data effectively.