Research

What we're working on

We work on the hard retrieval and modeling problems that need to be solved before AI can truly know individual users.

Understanding-based Retrieval

We're working on retrieval that understands the full intent of a query and can distinguish subtle semantic details — going beyond surface-level similarity to find what's actually relevant.

Multi-hop Retrieval

A lot of useful context lives in connections rather than individual memories. Something you mentioned two weeks ago might only be relevant when combined with something from yesterday — and neither piece is obviously useful on its own. We work on retrieval systems that can follow these chains, pulling together information that needs to be linked rather than just looked up.

Long-term User Modeling

The goal isn't just to store what happened — it's to build a model of who someone is. Their preferences, how they think, what they're trying to accomplish. Something that gets more accurate the longer you interact with it, not just a longer log of past conversations.