In 2000, Gabriel Kreiman published a Nature paper showing that single neurons in the human brain fire both when a person sees an image and when they imagine it. Co-authored with Christof Koch at Caltech, this was a landmark in memory neuroscience.
Twenty-six years later, Kreiman, then a professor at Harvard Medical School and Boston Children’s Hospital and associate director of the MIT-Harvard Centre for Brains, Minds and Machines, left academia to commercialise his research on how the brain encodes memory.
Kreiman’s startup, Engramme, founded in 2025 with co-founder Spandan Madan, emerged from stealth last month and is seeking to raise about $100 million, according to Bloomberg. Investors have discussed a potential valuation up to $1 billion, though terms are not final.
Previously, the company raised a $3 million pre-seed round led by Mayfield Fund and other Silicon Valley investors.
Engramme is developing Large Memory Models, a new class of AI systems designed to store and retrieve information more like the human brain does than traditional databases or search indexes do.
The architecture focuses on three key properties: lifelong storage at petabyte scale, proactive retrieval that surfaces relevant information without a search query, and associative recall that connects information across time and context, similar to how the hippocampus links related memories.
The company’s consumer iOS app is in beta, and an enterprise API for memory extraction and retrieval is in development. Through discussions with over 50 potential users, including older adults with memory loss, project managers, and AI developers, the team identified both a consumer market and a clear enterprise use case: preserving organisational knowledge often lost when employees leave.
Direct competitors include Mem0, Rewind AI, Zep, LangMem, and MemGPT. Engramme claims its approach is grounded in neuroscience rather than engineering convention, arguing that building memory systems based on how the brain works yields fundamentally different and better results than using vector databases and embedding models. This is a testable claim, but the company has not yet provided independent benchmarks.
If completed at a $1 billion valuation, the $100 million raise would be an aggressive pre-product entry for a company with only a $3 million pre-seed and a beta app. Investors must decide whether Kreiman’s two decades of peer-reviewed memory research provide a true competitive advantage or just a compelling narrative.
