Reasoning and judgment
We are interested in how models structure ambiguity, compare tradeoffs, maintain long-horizon objectives, and produce decisions that remain inspectable.
Evidence and synthesis
Agatha explores retrieval, source authority, contradiction handling, evidence sufficiency, citation integrity, and the boundary between answer generation and verified knowledge.
Memory and continuity
Useful continuity requires more than storing transcripts. We are developing explicit, scoped, versioned memory systems that preserve evidence and user control.
Interface intelligence
Different kinds of work require different interfaces. We study how conversation, editorial artifacts, visual prototypes, structured workflows, and proactive notices shape useful collaboration with AI.
Design-aware generation
Generative software and visual systems should preserve coherent design languages, responsive behavior, interaction states, and approved visual contracts.
Safe execution
We study architectures in which models can propose plans, tools, and mutations while deterministic systems retain authority over what runs and what counts as verified.