Research Lab

Original research at the boundaries of graphs, agents, and formally-verified systems

Productized Research

  1. 2019

    hCore: High-performance agent-based simulation engine developed in Rust, compiled to WASM, capable of running user-created JavaScript, Python and Rust simulations.
  2. 2019

    GPT-2 ABMs: Early LLM-integrated agent-based simulations, and grounding of agents in world models begins.
  3. 2020

    Real-world applications: Optimization of real-world systems like vaccine distribution and supply chains begins.
  4. 2021

    Graph-backed models: A multi-temporal, strongly-typed graph for For error-sensitive, safety-critical environments.

Active Research Areas

  • In-Transformer Simulation

  • Graph Machine Learning / JEPA

  • Temporal Hypergraph Querying

  • Autoformalization

  • Logic Programming

  • Process (De)composition

  • Hierarchical Reinforcement Learning

Recent Work

  • 2025 – ongoing

    Formally-verifiable modeling of safety-critical supply chains

    In collaboration with the University of Oxford’s Department of Computer Science: Professor Nobuko Yoshida, Professor David Parker, and Adrian Puerto Aubel. Read more on our blog.
  • 2025 – ongoing

    Petri Nets for trustworthy agentic coordination

    Exploring how Petri Nets, a mathematical formalism for modeling distributed, concurrent systems, can be used to coordinate and audit agentic workflows. Petrinaut is one of the outputs of this work.
  • 2025 – ongoing

    Specification elicitation tools for faithful intent representation

    One of several projects we’re working on as part of ARIA’s Safeguarded AI programme, where we lead the “Interaction Paradigms” / HCI technical area. Brunch is one of the outputs of this work.
  • 2024 – ongoing

    Multi-temporal hypergraph query language

    Designing a query language with first-class support for hypergraph relations and multiple, independent time dimensions — enabling precise reasoning over how data evolves and when it was known. HashQL is the output of this work.
  • 2026 – ongoing

    Trustable generative UIs

    Can we develop a design system and component library that can be safely used for generative UIs, with formal guarantees that the resulting interfaces are free from dark patterns — whether as a result of naive oversight or malign intent? Learn more at hash.design.
  • 2021 – ongoing

    SemType

    A shared, open registry of semantic types — machine-readable definitions of the concepts, measures, and entities that structured data refers to. SemType provides a common vocabulary that both people and AI agents can reference to exchange data unambiguously.
  • 2021 – 2024

    Embedder-agnostic blocks

    A research effort to make interactive UI blocks portable across applications, regardless of the host environment. The output of this work was the Block Protocol — an open standard for structured, composable blocks that can be embedded in any supporting application.
You can learn more about various of our projects on our developer site at hash.dev and design site at hash.design.

Join Us

We’re recruiting researchers to join our team to work on the projects above and more. Browse our open research roles below, or view all open roles.

Create a free account

Sign up to try HASH out for yourself, and see what all the fuss is about

By signing up you agree to our terms and conditions and privacy policy