Research
Research themes and analysis for durable AI systems.
What we study
Questions and commitments that shape how we build durable, inspectable AI systems for real operators.
Additional context for this page is coming soon.
Research themes
Persistent threads across our research, engineering, and open-source work.
Systems · structure
Explicit graphs and contracts
Model entities, relationships, and policies as first-class structure so automated reasoning stays inspectable—rather than burying logic in opaque prompt chains.
Domains · workflows
Operator-ready AI
Finance, marketing, and compliance behave like interconnected systems in real businesses. We study how to encode that reality so tools fit how people actually work.
Trust · documents
Documents, structure, and provenance
Long-form content needs parsing, segmentation, and traceable lineage when models consume it—primitives that hold up under audit and iteration.
Quality · operations
Evaluation and repeatability
One-off demos age quickly. We prioritize measurement harnesses, regression signals, and playbooks teams can run again when data and regulations shift.
Labs · code
Research that ships as artifacts
Memos propose abstractions; repositories and CLIs make them testable. Labs work is how we stress-test assumptions before they harden into product.
Work on the business. Not just in it.
HatTrickHQ helps operators drive consistent wins across finance, marketing, and compliance—without adding complexity.