Machine Learning Engineer

Mach Industries•Huntington Beach, CA

14d•$120,000 - $160,000

About The Position

Mach Industries is building an AI-forward autonomy stack for contested environments where GPS and other sensing are unavailable or unreliable. As a Machine Learning Engineer, you will own and scale the training, data, and edge-inference backbone that every vision and multi-sensor model on our product lines depends on for detection, tracking, search, navigation, targeting, and automatic target recognition. This is a broad, high-ownership role: you'll stand up the data and training infrastructure that lets the autonomy team iterate fast, generate synthetic data to cover the long tail, and get research-grade models running in real time on embedded hardware in flight. We are generalists, so you'll move fluidly between infrastructure, modeling, and deployment.

Requirements

Strong generalist software engineering: Python for ML and tooling, plus production C++ on Linux; profiling, optimization, and rigorous testing discipline.
Proven experience building ML data and training pipelines end to end: dataset construction, labeling/QA, augmentation, experiment tracking, and reproducible training.
Hands-on training and fine-tuning in PyTorch across modern detection/segmentation/tracking architectures (CNN/Transformer).
Edge and real-time deployment: model compression (INT8/FP16), runtime optimization (TensorRT/ONNX Runtime), and meeting latency/SWaP constraints on embedded GPU (Jetson-class) hardware.
Data and MLOps infrastructure: SQL/Parquet, dataset/versioning tools, CI-based validation, and scalable multi-GPU training.
BS/MS/PhD in CS/EE/Robotics or similar, or equivalent experience, with a track record shipping ML models to production or hardware. Senior candidates: deeper ownership of training/data infrastructure at scale.

Nice To Haves

Synthetic data generation and simulation (e.g. Unreal/Isaac, domain randomization) and demonstrated sim-to-real transfer.
EO/IR imagery experience and working with real flight/test data in challenging, degraded, or contested environments.
Multi-modal perception and fusion (EO/IR + radar/LiDAR/RF) at the feature or decision level.
Detection/tracking/search at scale; active learning and data-mining strategies for long-tail coverage.
CUDA backends for performance debugging; ROS 2; NVIDIA Jetson deployment pipelines.
Drift/dataset-shift monitoring, robustness and rare-event testing, long-horizon reliability metrics.
Distributed training frameworks and cloud ML platforms (e.g. SageMaker); Docker for reproducibility; Rust for systems tooling.

Responsibilities

Own and evolve the training and data infrastructure the autonomy team builds on: ingestion from flight/sim/HITL, curation and mining, labeling/QA workflows, dataset versioning (DVC/Parquet), and reproducible dataset builds.
Stand up and scale training/eval infrastructure: distributed multi-GPU training, experiment tracking, a model registry, and CI-based evaluation with regression gates plus automated field-data to retrain to validate to redeploy loops.
Deploy and optimize models for real-time edge inference on Jetson-class hardware (quantization/pruning, TensorRT/ONNX Runtime); profile CPU/GPU and hit tight latency, throughput, and SWaP targets.
Build and improve models across the portfolio as a hands-on IC: detection, segmentation, tracking, target/area search, classification/ATR, and multi-sensor fusion for EO/IR and auxiliary sensing.
Generate and manage synthetic data at scale (simulation + domain randomization) to cover long-tail and degraded conditions and close sim-to-real gaps.
Instrument runtime health, drift detection, and graceful degradation, and wire model-performance metrics back into the data and retraining loop.
Live close to flight data with visualization, triage, and root-cause tooling so the team can go from field logs to insight and model updates rapidly.
Partner with other autonomy disciplines across perception, localization, embedded, and flight-test to take capabilities from prototype to sim to HITL to flight to deployment.