Shawn Shen, the founder of Memories AI, asserts that for artificial intelligence (AI) to thrive in the physical realm, it must be capable of remembering what it perceives. His company is at the forefront of developing a visual memory layer tailored for wearables and robotics, utilizing advanced tools from Nvidia.
During the recent GTC conference, Memories AI announced a strategic collaboration with Nvidia. This partnership enables Memories AI to leverage Nvidia’s Cosmos-Reason 2, an innovative reasoning vision language model, alongside Nvidia Metropolis, a sophisticated reference architecture designed for video search and summarization. These technologies are pivotal for advancing the company's visual memory capabilities.
Shen, alongside his co-founder and CTO Ben Zhou, conceived the idea for Memories AI while working on the AI system for Meta’s Ray-Ban glasses. Their experiences with these smart glasses prompted them to consider the practical implications of AI technology in everyday life, particularly the necessity for users to recall the recorded video data.
Upon investigating existing solutions, they found a gap in the market for a visual memory solution that could effectively serve AI applications. Unable to find a suitable alternative, they opted to separate from Meta and embark on creating their own solution.
Shen highlighted, “AI is already excelling in the digital realm. But what about its capabilities in the physical world? AI wearables and robotics require memories too. We envision a future where AI has visual memories.”
The concept of memory in AI is relatively nascent. OpenAI began enhancing ChatGPT’s capabilities to remember previous chats in 2024, refining this feature into 2025. Additionally, companies like Elon Musk’s xAI and Google Gemini have introduced their own memory tools over the past two years. However, Shen points out that most of these advancements have concentrated on text-based memory, which, while structured and easily indexed, falls short for physical AI applications that predominantly engage with the environment through visual data.
Founded in 2024, Memories AI has successfully raised $16 million to date, comprised of an $8 million seed round in July 2025, followed by an additional $8 million extension. This funding round saw the participation of prominent investors including Susa Ventures, Seedcamp, Fusion Fund, and Crane Venture Partners, among others.
Shen explained that establishing a robust visual memory layer necessitates two critical components: the infrastructure to embed and index video data into a retrievable format, and the collection of sufficient data to train the model effectively. In July 2025, the company unveiled its large visual memory model (LVMM), which Shen likened to a more compact version of the recently launched Gemini Embedding 2, a model designed for multimodal indexing and retrieval.
To gather data, Memories AI developed a unique hardware device known as LUCI, which is worn by the company’s data collectors to record video for training purposes. Although they do not intend to become a hardware manufacturer or sell these devices, they created LUCI due to dissatisfaction with existing video recorders that prioritize high-definition quality and excessive battery consumption.
The company has since released the second generation of the LVMM and has entered into a partnership with Qualcomm to utilize their processors later this year. While Shen indicated that Memories AI is already collaborating with several major wearable manufacturers, he refrained from disclosing specific names. He believes that despite current demand, even greater opportunities within the wearables and robotics sectors lie ahead.
“Our focus in commercialization is primarily on the model and infrastructure. We are confident that the wearables and robotics market will evolve, but it may take time,” Shen remarked.
Source: TechCrunch News