Artificial intelligence is breaking through the final boundaries of video understanding. While current AI tools can analyze individual videos and generate summaries, they struggle when faced with thousands of hours of multi-video content. This technical bottleneck is troubling security companies and marketing enterprises, as the former need AI to screen massive surveillance footage, while the latter want to analyze different video marketing campaigns and product shooting materials.

A startup called Memories.ai is redefining this field with breakthrough technology. This AI platform has the impressive capability to process up to 10 million hours of video, offering enterprises with large amounts of video data a complete context understanding layer, including searchable indexes, a tagging system, clip segmentation, and data aggregation features.

Investment, funding, money

The company's two co-founders have deep backgrounds. Dr. Shen previously worked as a research scientist at Meta's Reality Labs and earned his doctorate, while Zhou Wenmin served as a machine learning engineer at Meta. This technological foundation laid a solid base for Memories.ai's innovation.

Dr. Shen pointed out the core issue in the current AI field during an interview: "Top AI companies like Google, OpenAI, and Meta focus on developing end-to-end models. These capabilities are indeed excellent, but these models often have limitations in understanding video contexts beyond one or two hours."

He further explained: "However, when humans use visual memory, we filter through large amounts of contextual data. Inspired by this, we hope to build a solution that better understands video content spanning several hours."

This vision has received strong recognition from investors. Memories.ai has just completed an $8 million seed round led by Susa Ventures, with participation from Samsung Next, Fusion Fund, Crane Ventures, Seedcamp, and Creator Ventures. Notably, the company initially aimed to raise $4 million, but due to strong investor interest, it ended up oversubscribing.

Misha Gordon-Ro, a partner at Susa Ventures, expressed high praise for the founders: "Shen is a technically strong founder who is obsessed with pushing the boundaries of video understanding and intelligence. Memories.ai can unlock a large amount of first-party visual intelligence data through its solutions. We see a gap in long-context visual intelligence in the market, which is exactly why we invested in this company."

Samsung Next's investment logic differs. The Samsung investment division saw potential in Memories.ai's consumer market. Sam Campbell, a partner at Samsung Next, said: "What we value about Memories.ai is its ability to perform extensive device-side computing. This means you don't necessarily need to store video data in the cloud. It can provide better security applications for people who are concerned about privacy and reluctant to install home security cameras."

The technical architecture of Memories.ai showcases its innovative strength. The company uses its own developed technology stack and models for analysis, first removing noise from videos, then processing them through a compression layer to store only important data. Next is the indexing layer, which allows video data to be searched via natural language queries and provides segmentation and tagging features. Finally, the aggregation layer summarizes the indexed data to help generate reports.

Currently, the startup primarily serves two types of enterprises: marketing companies and security companies. Marketing companies can use its tools to find trends related to their brands on social media and determine the types of videos to produce. Memories.ai also provides video creation tools for marketers. At the same time, the company is collaborating with security companies to help them analyze surveillance footage, identifying potential dangerous behaviors of people in videos through pattern reasoning.

At this stage, enterprises partnering with Memories.ai need to upload their video libraries to the platform for analysis. However, Dr. Shen revealed that in the future, customers will be able to create shared drives and more easily synchronize content. The ultimate goal is for users to be able to ask questions such as "Tell me about everyone I interviewed last week."

Dr. Shen's vision is even grander. He envisions developing an AI assistant that can gain contextual understanding of a user's life through photos or smart glasses. He also believes this technology will play an important role in training humanoid robots to perform complex tasks or helping autonomous vehicles remember different routes.

The company currently has 15 employees and plans to use the newly raised funds to expand the team and improve search functions. In terms of competition, Memories.ai faces similar startups such as mem0 and Letta, which are also providing memory layers for AI models, although their current video support is limited. Additionally, it must compete with companies like TwelveLabs and Google, which have already made progress in helping AI models understand videos.

However, Dr. Shen believes their solution is more horizontally scalable and can work collaboratively with different video models. This technological advantage could be the key factor that sets Memories.ai apart in a competitive environment.

As video content explodes in the digital world, AI technologies capable of effectively understanding and analyzing long, multi-source video content will become the next technological frontier. Memories.ai's breakthrough not only provides enterprises with powerful video analysis tools but also opens new possibilities for the development of artificial intelligence in the field of visual understanding.