Samsung Backs Video AI Startup Specializing in Analyzing Thousands of Hours of Footage

Many AI tools can analyze a video today and summarize its content, but things become challenging when models are queried about multiple videos and footage that extends over many hours.

This represents a significant limitation for security firms seeking to utilize AI to sift through thousands of hours of footage from various cameras, as well as for marketing agencies aiming to examine different video campaigns and product shoots.

Memories.ai aims to address this challenge with its AI platform, which can process up to 10 million hours of video. For companies with extensive video to analyze, the startup intends to provide a contextual layer, featuring searchable indexing, tagging, segmenting, and aggregation.

Its co-founder, Dr. Shawn Shen, previously worked as a research scientist at Meta’s Reality Labs while pursuing his PhD. His counterpart, Enmin (Ben) Zhou, served as a machine learning engineer at Meta.

“All top AI companies, such as Google, OpenAI, and Meta, focus on producing end-to-end models. While these capabilities are impressive, they often struggle to grasp video context beyond one or two hours,” Shen told TechCrunch.

“However, humans utilize visual memory by sifting through large datasets. This inspired us to create a solution that enhances the understanding of video across extended timeframes,” he explained.

Image Credits: Memories.ai

To this end, the company has recently secured $8 million in a seed funding round led by Susa Ventures, with participation from Samsung Next, Fusion Fund, Crane Ventures, Seedcamp, and Creator Ventures. Shen mentioned that the original goal was to raise $4 million, but the round was oversubscribed due to strong investor interest.

“Shen is a highly technical founder, and he is passionate about advancing the frontiers of video understanding and intelligence,” stated Misha Gordon-Rowe, a partner at Susa Ventures. “Memories.ai has the potential to unlock significant first-party visual intelligence data with its solution. We identified a market gap for long-context visual intelligence, which motivated our investment in the company,” he added.

Samsung Next approached the investment with a slightly different perspective: the investment arm of Samsung sees Memories.ai’s solution as valuable for consumers.

“One aspect we appreciated about Memories.ai is its capability for substantial on-device computing. This means there’s no strict need to store video data in the cloud, potentially unlocking better security applications for individuals wary of installing security cameras in their homes due to privacy issues,” remarked Sam Campbell, a partner at Samsung Next.

Memories.ai claims it leverages its proprietary tech stack and models for analysis. Initially, it eliminates noise from videos and feeds the output through a compression layer to retain only essential data. Following this, an indexing layer is applied to make video data searchable (using natural language queries), along with segmentation and tagging. Additionally, an aggregation layer summarizes data from the index to assist in report creation.

At present, the startup serves two primary types of clients: marketing and security. Marketing companies can utilize the startup’s tools to research trends related to their brands on social media and determine the types of videos they wish to create. Memories.ai also offers tools for marketers to produce those videos.

The company collaborates with security firms to analyze footage and identify potentially hazardous behaviors by evaluating patterns in the videos.

Image Credits: Screenshot by TechCrunch

Currently, companies partnering with Memories.ai need to upload their video libraries to the platform for analysis. However, Shen mentioned that in the future, clients will have the ability to establish a shared drive and synchronize content more seamlessly. The plan includes enabling customers to ask questions like, “Tell me all about the people I interviewed last week.”

Shen envisions an AI assistant that can contextualize a user’s life through their photos or when they activate smart glasses. He also believes the technology could play a role in training humanoid robots for complex tasks and helping self-driving cars retain different routes.

The company currently employs 15 individuals and plans to utilize the funding to expand its team and enhance its search capabilities.

Memories.ai faces competition from similar startups, such as mem0 and Letta, which are working on providing a memory layer for AI models, though they currently offer limited video support. Moreover, it must contend with firms like TwelveLabs and Google, both working toward improving AI models’ comprehension of videos.

However, Shen is confident that his company’s solution is more horizontal, allowing it to integrate with various video models as well.