Model Card for DeOSAlphaCompactWorldModels

Model Details

DeOSAlphaCompactWorldModel is a neuro-symbolic meta-model that integrates multiple compact world models designed for decision optimization within the DeOS (Decision Optimization System) framework by vencortex®.

It enables simulation and prediction of complex decision environments by orchestrating several domain-specific models—including goal, cost, and dynamics models—into a unified world representation.

Developed in the context of investment and other high-stakes business decisions, DeOSAlphaCompactWorldModel serves as the reasoning component of multimodal decision optimization. It enables training and acting within a simulated world to achieve complex business outcomes under contextual constraints and strategic goals.

State of Research

DeOSAlphaCompactWorldModel was initially developed for investment and other high-stakes business decisions, but it is designed to be generalizable to a wide range of decision-making domains, such as medical, operational, or strategic planning applications.

The goal is to provide a trainable framework and eventually open-source components to enable researchers and practitioners to adapt and extend the model to domain-specific small world models.

A detailed research paper describing the architecture, implementation, and evaluation is currently in preparation. If you are interested in contributing or extending the model for specific domains, please reach out to collaborate.

Example Usage

# Requires: deosalpha >= 2025.10.12
# Copyright (c) 2025. vencortex® - DeOS | Decision Optimization System. All rights reserved.

from deos_alpha.capabilities.world_models.predictor import WorldModelConfig, DeOSAlphaWorldModelPredictor
from deos_alpha.infrastructure.environment_reader import EnvironmentReader

# Define model configuration
config = WorldModelConfig(
    domain_model_id="example_domain_model",
    goal_model_id="test_goal_model",
    cost_model_id="example_cost_model",
    dynamics_model_id="example_dynamics_model",
    data="test_dataset",  # optional
    api_key=EnvironmentReader().get_deosalpha_api_key()
)

# Initialize predictor
predictor = DeOSAlphaWorldModelPredictor(config=config)

# Run simulation or prediction
simulation = predictor.predict(
    action=<Tensor>,      # your action input
    time_horizon=1        # number of simulation steps
)