Bayesian Phd Jobs in Michigan (NOW HIRING)

Role: Automotive Propulsion Software Research Engineer
Location: Fully onsite. Warren, MI
Contract role
Job description:
Develop an LLM and agent based assistant to automate simulation setup, run deck generation, boundary condition/mesh recommendations, results review, and requirements alignment across 1D/3D tools (e.g., GT POWER/Simcenter Amesim, ANSYS/STAR CCM+, MATLAB/Simulink).
Surrogate Modeling & Automated Design Space Exploration (DSE): Build physics informed ML surrogates for key propulsion performance metrics (e.g., torque, BSFC/efficiency, thermal limits, emissions proxies, e drive efficiency maps) and integrate them with active learning data generation and multi objective optimization
MS or PHD in Computer Science, Computer Engineering, or Electrical Engineering
Skillset: Python programming (critical); Experience in AI/ML, preferably Generative AI (critical); familiarity with embedded software development lifecycle and systems engineering activities
• System architecture covering data sources (requirements, prior studies, PLM), tools (1D/3D solvers), and results storage (Lakehouse).
• Document parsing pipelines for requirements/specs/test plans; establish content normalization (Markdown/JSON).
• Agent workflows for: interpreting change requests, proposing simulation plans, generating run decks/templates, mesh/BC suggestions, and automated post processing.
• Chat + form based UI to request studies, generate inputs, and review outputs.
• Integration with Databricks (Lakehouse, MLflow, Model Serving) for versioning, observability, and scalable compute.
• Define target responses (e.g., torque map, efficiency map, pressure ratios, temperatures, emissions proxies), inputs (geometry/controls/operating points), and constraints.
• Data model and schema for simulation inputs/outputs; provenance and UQ metadata.
• Orchestrate solver sweeps (1D/CFD/FEA where applicable) to seed the dataset.
• Implement adaptive sampling/active learning to target high value points that reduce model error.
• Train candidate surrogates (e.g., GPR, XGBoost/TabNet, feedforward NNs, physics informed NNs).
• Cross validation, error budgets, calibration; uncertainty quantification and guardbanding.
• Integrate surrogates with DSE (e.g., Bayesian Optimization, NSGA II) to target objectives (efficiency/BSFC, mass, cost, thermal margins) under constraints (emissions/temperature/packaging).
• Optional coupling to MDAO toolchains (e.g., ModelCenter/HEEDS/Simcenter).