Emerging Visual Memory Technology in AI
Shawn Shen emphasizes that for artificial intelligence to thrive in the physical world, it must possess the ability to remember what it observes. His company, Memories.ai, is leveraging Nvidia’s advanced AI tools to create an infrastructure that allows wearables and robotics to recall and utilize visual memories effectively.
Strategic Collaboration with Nvidia
Memories.ai recently unveiled a partnership with Nvidia during its GTC conference. This collaboration harnesses Nvidia’s Cosmos Reason 2, an advanced reasoning vision language model, alongside Nvidia Metropolis, which specializes in video search and summarization. Together, these tools will enhance the development of the company’s visual memory technology.
Founders’ Vision from Previous Experience
Shen shared with TechCrunch that he and co-founder and CTO Ben Zhou conceptualized the company while developing the AI system for Meta’s Ray-Ban glasses. Their work led them to contemplate how users would interact with such technology if they were unable to recall the video data they captured.
Identifying a Market Gap
Upon realizing that no existing solutions addressed a visual memory system for AI, they decided to leave Meta and forge their own path. Their research confirmed a significant opportunity: building a solution that enables AI wearables and robotics to have memories akin to human-like recall.
The Importance of Visual Memory in AI
Shen pointed out that while AI has advanced considerably in the digital landscape, its capabilities in the physical world still require substantial development. He believes that for AI to function effectively in environments filled with visual stimuli, it must be equipped with visual memory.
Investment and Growth Trajectory
Founded in 2024, Memories.ai has successfully raised $16 million to date. This funding includes an $8 million seed round completed in July 2025, with contributions from prominent investors such as Susa Ventures, Seedcamp, Fusion Fund, and Crane Venture Partners.
Innovative Data Collection Approach
To facilitate the creation of its visual memory layer, Shen highlighted the need for robust infrastructure to embed and index videos in a retrievable format, as well as comprehensive data gathering to train the model. The company launched its large visual memory model (LVMM) in July 2025, designed to compete with multimodal indexing models like Gemini Embedding 2.
Future Aspirations in Wearables and Robotics
Memories.ai employs a dedicated device called LUCI, worn by its “data collectors,” to capture video footage needed for model training. Although Shen asserts that the company does not aim to transition into hardware production, he recognized the limitations of existing high-definition video recorders. He intimated that while they are currently collaborating with notable players in the wearable sector, he anticipates even greater opportunities in wearables and robotics on the horizon.
As Shen noted, the company’s focus remains on refining their model and infrastructure, with a belief that the wearable and robotics markets will expand significantly in the near future.