About

Research in simulation, world models, and environment dynamics.

Applied Worlds publishes original, empirical work on how environments evolve, how to simulate them faithfully, and how agents learn and decide inside complex systems.

Financial markets, multi-agent environments, and physical systems under adaptive control share a common structure: they are non-stationary, partially observable, and endogenously shaped by the agents acting inside them. Building computational representations that are good enough to learn from, plan with, and decide inside is the problem all three research threads address.

Each artifact follows the same structure: a hypothesis, an experiment, an implementation, and a published record of results — including failures. Negative results are published alongside positive ones.

Applied Worlds is the Systems leg of a three-site research ecosystem. It connects to Applied Markets — how agents behave in financial environments — and Applied Models — how neural representations form internally. The three questions are complementary: Systems × Behavior × Intelligence.

Research Threads Three
01
Thread I

World Models

Learning a latent representation of how a system evolves — then planning in that space without acting in the real environment. Covers learned dynamics models, physics-informed hybrid architectures that combine known structure with learned components, and model-based planning inside simulated environments.

Latent Dynamics Dreamer Physics-Informed Neural Networks State-Space Models Model-Based Planning
02
Thread II

Agent-Environment Dynamics

When an agent acts, the environment changes — which changes the agent's next observation, which shapes the agent's next action. In markets this is reflexivity. In multi-agent RL this is non-stationarity — every agent's policy is part of every other agent's environment. In macroeconomics this is the Lucas critique. Experiments involve building simulated environments with multiple adaptive agents and studying what emerges from the feedback loop.

Reflexivity Multi-Agent RL Non-Stationarity Lucas Critique Emergence
03
Thread III

Counterfactual Simulation

Simulation is only useful if you can intervene inside it, not just run it forward. Structural causal models provide the formal language for "what if?" — and when encoded as simulable environments, they allow measurement of whether counterfactual reasoning improves decisions compared to purely observational baselines.

Structural Causal Models Do-Calculus Intervention Counterfactual Reasoning Decision Quality
What Gets Published Artifact types
Experiments
Computational Studies

Implementation, training run, evaluation, results. Reproducible where possible, honest about limitations always.

Articles
Analytical Pieces

Longer-form writing where an experiment alone does not tell the full story. Always grounded in results.

Notebooks
Interactive Code

Executable walkthroughs of methods and implementations. Not tutorials — structured explorations.

Always
Failures Included

Negative results ship alongside positive ones. The record of what was tried is as valuable as what worked.

Editorial Boundary Rules
No Learning Notes Not passive study output. Course notes, reading reflections, and literature summaries do not belong here.
No Reposting No paper summaries, no blog recaps. Every artifact is original work built on direct implementation.
No Tutorials Applied Worlds is not a teaching resource. Every artifact starts with a hypothesis, not an explanation.
Failures Published If the experiment does not work, that is published too. The record is the product.
Keep Moving Progress over optics. There is no content calendar. Artifacts ship when the experiment is done.
Behind the Work

Applied Worlds is a personal initiative. Other projects from the same desk: Prachalabs.com — products and tools, including Rewire — and Pracha.me — personal site and broader writing.