Today’s AI models are fluent and capable, yet fundamentally impersonal. By default, they give the exact same answer to anyone who asks. We believe personalization is what unlocks the next order of magnitude in AI value: the shift from passive tools that people use, to proactive, model-centric systems that self-evolve to suit their needs. In our future, AI won’t just know the world; it will actually know you.
How do you compress a person's preferences, communication style, and intent into a form a language model can act on? We study both natural language paths — search, retrieval, dynamic prompting — and embedding paths — persona vectors, cross-attention conditioning — as orthogonal, complementary mechanisms.
Every conversation is a signal. We develop methods for dense, per-turn reward estimation — predicting user behavior alongside generation, then comparing prediction against reality. Models that improve continuously from interaction, not just from curated data.
In most systems, the model generates text at the end of a pipeline it doesn’t control — memory, personalization, content selection, and interface are all orchestrated by code, with the model as one component among many. We put the model at the center. It controls what it remembers, how it probes and learns about a user, what content to surface, and how the product behaves. When these capabilities live inside a single model rather than separate modules, they compound — and the system becomes end-to-end optimizable in ways that pipeline architectures cannot be.
Offline metrics routinely overestimate online impact by an order of magnitude. We build replay-based benchmarks, automated experiment pipelines, and causal measurement that bridges that gap.
A single reply is solved. The hard problem is what happens over time — models contribute no new information of their own; they recombine what creators and users provide until the well runs dry. We study how to sustain information density across long-running content: models that introduce genuine novelty, structure, and surprise on their own, not just when prompted.
Millions of users interact with our product every day. We build systems that treat this collective activity as a living knowledge base — extracting patterns from how users engage, what creators invent, what succeeds and fails — and feeding it back into the model as a continuous loop, not static training data.
We are researchers and engineers who ship. Every research direction connects to a live product serving millions of users. We validate ideas through rapid experimentation, not just paper benchmarks — and we publish what we learn.
AI is becoming the primary interface between people and information, entertainment, education, and work. If these systems cannot adapt to individuals, they will flattenhuman experience rather than enrich it. Getting personalization right isn't a product optimization — it's a prerequisite for AI that genuinely serves people.
We're Kaon. We're here to make AI personalized.