← Back to All Openings

Staff Systems Engineer – Joint Belief‑World Model Autoregressive Architecture

corpora-jobs-1778796293285-db9d41c6 - Frontier Development
Systems EngineerStaff1 position
⚡

Why This Role is Different

Frontier Development Role

You will build the core predictive engine that lets agents forecast their own future beliefs and observations, turning the BAAC framework from a conceptual design into a deployable system that runs at real‑time rates on edge devices.

The Frontier Element

By fusing belief and observation prediction in a single autoregressive loop, you will create the first model that can ‘imagine the next view’ while simultaneously predicting the next action—an ability that has never been realized at scale in multi‑agent settings.

🔬

Project Context

Research Area

Joint Belief‑World Model (JBWM)

From: Partial Observability Amplification of Misalignment

Why This Role is Critical

To design, implement, and optimize a unified autoregressive model that predicts both the next observation and the next belief vector, enabling agents to interleave imagination and action prediction for real‑time misalignment correction.

What You Will Build

A scalable transformer‑based JBWM, end‑to‑end training pipeline, low‑latency inference engine, and integration hooks for the belief hierarchy and communication modules.

🛠

Key Responsibilities

  • Architect a transformer‑based autoregressive model that jointly predicts belief vectors and observations conditioned on past actions and communicated beliefs.
  • Optimize the model for 10 Hz inference on GPU‑edge hardware, including mixed‑precision and model‑parallelism techniques.
  • Develop a training framework that alternates between imagination and action prediction, ensuring stability and low compounding error.
  • Integrate the JBWM with the belief hierarchy and communication modules, exposing joint belief estimates to downstream reward‑decomposition logic.
  • Benchmark performance on SMAC, MPE, and UAV swarm tasks, publishing results and open‑source the code.
🎯

Required Skills & Experience

Technical Must-Haves

Transformer Architectures and Autoregressive Modeling

Expert
Designing large‑scale sequence models that predict multi‑modal outputs.

CUDA Programming and Mixed‑Precision Training

Advanced
Achieving real‑time inference on edge GPUs.

Deep RL Training Pipelines

Advanced
Integrating policy gradients with autoregressive world‑models.

Software Engineering (C++, Python, Docker)

Proficient
Building production‑grade inference services.

Experience Requirements

  • 4+ years of systems engineering for large ML models, with a portfolio of deployed inference engines.
  • Published work on transformer‑based autoregressive models or diffusion models.

Education

PhD or Master’s in Computer Science, Electrical Engineering, or a related field with a focus on deep learning systems.

⭐

Preferred Skills

  • Experience with diffusion models or graph‑based attention.
  • Knowledge of multi‑agent simulation environments (SMAC, MPE, UAV swarms).
🤝

You Will Thrive Here If...

  • Comfortable shipping code in production before formal reviews.
  • Strong bias toward action and rapid iteration.
📈

Impact & Growth

12-Month Impact

Within 12 months, deliver a JBWM inference engine that runs at ≥10 Hz on a single edge GPU, reduces misalignment by 25 % on benchmark tasks, and is integrated into the BAAC product pipeline.

Growth Opportunity

Lead the system architecture team, mentor junior engineers, and guide the transition from prototype to scalable, cloud‑native deployment.

Ready to Push the Boundaries?

If this sounds like the challenge you have been looking for, we want to hear from you. We value what you can build over where you have been.