Generative Modeling Engineer

About Company
Data science company with a software platform that runs secure and compliant machine learning workloads at scale. We specialize in auditable environments and in synthetic data for defense use cases where the source data cannot leave a government installation. Our first program builds physics-informed failure data for aircraft drivetrain health monitoring, working with an engineering partner that models how bearings and gears fail.

THE ROLE
You will build the generative models that challenge our physics-first approach to synthetic vibration data. Working with our data scientist, you will train conditional diffusion models on physics-generated fault signals and sensor spectra, and test them against a domain adaptation baseline on held-out data. The work runs inside a fixed GPU budget in AWS GovCloud, so every training run is planned and every result has to stand up to a skeptical engineering review.

WHAT YOU WILL DO
▪ Train conditional diffusion models (DDPM or latent diffusion) on vibration time series, spectrograms or envelope spectra, conditioned on operating regime and fault severity.
▪Build the physics checks every generated sample must pass, including defect-frequency content and amplitude statistics.
▪Compare generators against a feature-level domain adaptation control on held-out data, split by source.
▪Run memorization and membership-inference audits on trained models, and test differentially private training.
▪Work inside fixed monthly GPU run budgets with checkpointing, early stopping and a full experiment record.
▪Document methods, results and limitations for the program reports.

REQUIRED
▪ Four or more years building deep generative models in PyTorch, with hands-on diffusion model work.
▪ Experience with time series, audio, vibration or other signal data.
▪ Disciplined experiment tracking and evaluation on held-out data.
▪ Comfort working under a fixed compute budget on AWS or a comparable cloud.

PREFERRED
▪ Differential privacy (DP-SGD) or privacy audits of generative models.
▪ Physics-informed or physics-guided generative modeling.
▪ Amazon SageMaker training jobs and hyperparameter tuning.
▪ Rotating machinery or condition monitoring data.

WORKING ARRANGEMENT
This is a remote role, open to candidates anywhere in the United States. Company provides the hardware and a company-managed laptop.

CITIZENSHIP AND CLEARANCE
U.S. citizenship required. An active Secret clearance is a bonus; eligibility to obtain one required.

EEO:
“Mindlance is an Equal Opportunity Employer and does not discriminate in employment on the basis of – Minority/Gender/Disability/Religion/LGBTQI/Age/Veterans.”