Shawn Shen asserts that for AI to thrive in the physical realm, it must have the capability to remember visual information. His company, Memories.ai, is leveraging Nvidia’s AI technology to create a foundation for wearables and robotics to store and retrieve visual memories.
On Monday, Memories.ai revealed a partnership with semiconductor leader Nvidia at its GTC conference. This collaboration involves the use of Nvidia’s Cosmos-Reason 2, a reasoning vision language model, and Nvidia Metropolis, a tool for video search and summarization, to advance its visual memory solutions.
Shen, alongside his co-founder and CTO Ben Zhou, shared with JS that the inspiration for the company stemmed from their work on the AI system for Meta’s Ray-Ban glasses. They realized the potential for practical applications of such technology would be limited if users couldn’t recall the video data captured by the glasses.
After failing to find existing solutions for AI-based visual memory, Shen and Zhou decided to leave Meta and develop the technology independently.
“AI is excelling in the digital space, but what about the physical world?” Shen questioned. “AI wearables and robotics also require memory. We envision a future where AI possesses visual memories.”
The concept of AI systems having memory is relatively new. OpenAI updated ChatGPT to remember past interactions starting in 2024, refining this feature in 2025. Similarly, Elon Musk’s xAI and Google Gemini have introduced memory functionalities over the past two years.
However, Shen noted that these developments primarily target text-based memory, which is more structured and easier to organize but is less effective for AI applications that engage with the world visually.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
Memories.ai was established in 2024 and has secured $16 million in funding, comprising an $8 million seed round in July 2025 and an additional $8 million extension. The funding round was led by Susa Ventures, with participation from Seedcamp, Fusion Fund, and Crane Venture Partners, among others.
According to Shen, developing this visual memory layer required two key elements: creating the infrastructure to embed and index videos into a retrievable data format, and gathering the necessary data for training the model.
The company introduced its large visual memory model (LVMM) in July 2025. Shen compared it to a scaled-down version of Gemini Embedding 2, a newly released multimodal indexing and retrieval model.
For data collection, the company developed LUCI, a hardware device worn by “data collectors” that captures video to train the model. Shen stated that they do not intend to become a hardware company or sell these devices, but created their own because existing video recorders focused on high-definition formats that drained battery life.
The second generation of this LVMM has been launched, and the company has entered a partnership with Qualcomm to use Qualcomm’s processors starting later this year.
While Memories.ai is already collaborating with some major wearable companies, Shen chose not to disclose their identities. He believes that although demand is growing now, the real opportunities in wearables and robotics are yet to be realized.
“We are prioritizing the model and infrastructure development because we anticipate that the wearables and robotics market will eventually expand, but it might not be immediate,” Shen commented.
