Remote: Causal AI Research Intern Models Job Singapore,IT/Tech

Position: Remote: Causal AI Research Intern for World Models

We are seeking a deeply curious AI Research Intern to join our team in building and training the next generation of Causal AI models.

Company: Nubio.

World

Location: SINGAPORE / REMOTE

Reports to: Head of Research / CTO

About Nubio.

World:
Forging the Causal AI Frontier

At Nubio, we are architecting Causal AI Operating Systems to decode and master the hidden physics of high-stakes industries. We treat complex sectors—finance, aviation, energy, and supply chains—as intricate, simulatable universes.

Our platforms learn the true causal structure of complex systems—not correlations—and run 100,000+ "what-if" simulations to find optimal policies. We replace decades-old heuristics with physics-informed optimization.

We draw inspiration from Deep Mind's breakthroughs:
Alpha Go discovering "Move 37," Alpha Fold solving protein structure, Graph Cast revolutionizing weather prediction. We believe causal world models will create unprecedented value across industries where decisions propagate through networks under uncertainty.

The Role:
Prototyping Causal World Models

We are seeking a deeply curious AI Research Intern to join our team in building and training the next generation of Causal AI models. You will work alongside senior researchers to implement, test, and analyze models that learn the "physics" of complex industrial systems.

This role is focused on research and development. You will implement novel architectures, run large-scale JAX simulations, analyze training runs, and push the boundaries of causal discovery and policy optimization.

What You Will Do

Implement Causal Discovery Models: Build and test Differentiable Causal Discovery (DCD/NOTEARS) architectures in JAX/Flax that learn causal graphs from observational data:
Action → Outcome ← Competitor ← Exogenous Factors.
Build World Model Components: Implement and debug our Hierarchical World Model architecture:
- Slow Model (Cortex): Causal Transformer for strategic macro-trend forecasting
- Fast Model (Hippocampus): LSTM with boundary conditions for tactical real-time adaptation
Implement Policy Optimization: Build and test Constrained Markov Decision Process (CMDP) agents using Lagrangian relaxation for policy optimization under business constraints.
Diffusion Model Development: Train diffusion models that generate synthetic operational data for domains with limited historical records.
Data Analysis & Visualization: Analyze training runs—loss curves, causal graph convergence, policy performance. Visualize learned causal structures and identify failure modes.
Benchmark & Validate: Benchmark models against industry baselines and validate on real-world data from public sources.
Literature Review: Stay current on causal ML, world models, and RL research. Present key papers (NOTEARS, Genie, MuZero, Graph Cast) to the team.

Ideal Candidate Profile

Academic Background: Currently pursuing M.S. or Ph.D. (or advanced B.S. with research experience) in Computer Science, Machine Learning, Physics, Applied Mathematics, or related quantitative field.
Strong Coder: Deep proficiency in Python. Hands-on experience with JAX strongly preferred;
PyTorch experience acceptable with willingness to learn JAX.
Deep ML/AI Foundation: Strong understanding of:
- Deep learning fundamentals (back propagation, optimization, regularization)
- Reinforcement learning (policy gradient, actor-critic, MCTS)
- Probabilistic modeling (VAEs, diffusion models, Bayesian inference)
- Bonus:
  Causal inference (SCMs, do-calculus, instrumental variables)
First-Principles Thinker: Driven by curiosity to understand why things work. You read papers, not just tutorials. Comfortable deriving gradients by hand.
Data-Driven & Rigorous: Strong quantitative background. You design experiments properly, track metrics obsessively, and stress-test results before believing them.
Mission-Aligned: Excited by modeling the "physics" of complex systems—demand elasticity, competitive dynamics, network effects, stochastic optimization—not just achieving SOTA on benchmarks.

Research Areas You May Explore

Inverse RL for Competitor Modeling: Infer competitor objectives from observed behavior.
Physics Integration: Integrate external physics models (weather, network dynamics) into causal forecasting.
Multi-Modal Futures: Handle distributional uncertainty in long-horizon planning.
Causal Representation Learning: Learn causal structure directly from raw sequential data.

#J-18808-Ljbffr