Memories.ai is building the visual memory layer for wearables and robotics

Shawn Shen believes that AI volition request to retrieve what it sees successful bid to win successful the carnal world. Shen’s institution Memories.ai is utilizing Nvidia AI tools to physique the infrastructure for wearables and robotics to beryllium capable to retrieve and callback ocular memories.

Memories.ai announced a collaboration with semiconductor elephantine Nvidia astatine its GTC league connected Monday. Through this partnership, Memories.ai uses Nvidia’s Cosmos Reason 2, a reasoning imaginativeness connection model, and Nvidia Metropolis, an exertion for video hunt and summarization, to proceed to make its ocular representation technology.

Shen (pictured supra left) told TechCrunch that helium and his co-founder and CTO Ben Zhou (pictured supra right), got the thought for the institution portion gathering the AI strategy down Meta’s RayBan glasses. Building the AI glasses got them reasoning astir however radical would really usage the tech successful existent beingness if users couldn’t callback the video information they were recording.

They looked astir to spot if they could find anyone already gathering that benignant of ocular representation solution for AI. When they couldn’t, they decided to rotation retired of Meta and physique it themselves.

“AI is already doing truly good successful the integer world, what astir the carnal world?” Shen said. “AI wearables, robotics request memories arsenic well. … Ultimately, you request AI to person ocular memories. We judge successful that future.”

The quality for AI systems to remember, successful general, is simply a comparatively new. OpenAI updated ChatGPT to commencement to retrieve past chats successful 2024 and fine-tuned that diagnostic successful 2025. Elon Musk’s xAI and Google Gemini person launched their ain representation tools excessively successful the past 2 years.

But these advancements person mostly focused connected text-based memory, Shen said. Text-based representation is overmuch much structured and easier to scale but aren’t arsenic adjuvant for carnal AI applications that mostly interact with the satellite done show and visuals.

Memories.ai was launched successful 2024 and has raised $16 cardinal frankincense far, done an $8 cardinal effect circular successful July 2025 and an $8 cardinal extension. The circular was led by Susa Ventures and included Seedcamp, Fusion Fund and Crane Venture Partners, among others.

Shen said successfully gathering this ocular representation furniture required 2 things: gathering the infrastructure needed to embed and scale videos into a information format that tin beryllium stored and recalled, and capturing the information needed to bid the exemplary to bash conscionable that.

The institution launched its large ocular representation exemplary (LVMM) successful July 2025. Shen said it could beryllium compared to a smaller mentation of Gemini Embedding 2, a multimodal indexing and retrieving model, that was released earlier this month.

For information collection, the institution created LUCI, a hardware instrumentality worn by the company’s “data collectors” that records video utilized to bid the model. Shen said they don’t program to go a hardware company, nor merchantability these devices, but, rather, that they built their ain due to the fact that they weren’t satisfied with off-the-shelf video recorders that focused connected high-definition and battery-eating video formats.

The institution released the 2nd procreation of this LVMM and signed a partnership with Qualcomm to tally connected Qualcomm’s processors starting aboriginal this year.

Memories.ai is besides moving with immoderate of the ample wearable companies already, Shen said, but declined to disclose which ones. Despite immoderate request now, Shen sees adjacent bigger opportunities successful wearables and robotics yet to come.

“In presumption of commercialization, we are much focused connected the exemplary and the infrastructure, due to the fact that yet we deliberation the wearables and robotics marketplace volition come, but it’s astir apt conscionable not now,” Shen said.