ML Engineer — Edge Intelligence & Platform

WHO WE ARE
Small team. High trust. No bureaucracy. iTerra is building the Arsenal of the Caribbean — a sovereign-capable, dual-use autonomous detection and classification platform for defense and critical infrastructure. No tourists. Only builders.

WHAT YOU OWN

You own the intelligence layer. You build the models that detect what the system has never seen, classify what it has, and get smarter with every deployment. You work at the intersection of RF signal processing, machine learning, and agentic AI — turning raw spectrum data into autonomous threat intelligence at the edge. This is not a research role. You ship models that run on constrained hardware in contested environments.

• UAS classification models: Build and maintain the models that classify detected RF anomalies into threat categories — drone type, platform class, threat probability, and operator behavior. RF features in, actionable classification out. Your models are the product.
• RF feature engineering: Extract discriminative features from raw IQ data and spectral observations — modulation type, hop patterns, bandwidth, signal duration, burst characteristics, duty cycle, Doppler signatures. You define what the model sees and how it learns.
• Edge inference optimization: Deploy quantized, pruned, and optimized models on NVIDIA edge embedded compute under strict SWaP-C constraints. TensorRT, ONNX, TFLite — you own the path from trained model to production inference at the edge. Latency matters. Every millisecond counts.
• Training pipeline and MLOps: End-to-end pipeline from structured RF datasets → feature extraction → model training → validation → versioning → deployment. Automated retraining triggered by new field data. Reproducible experiments. Model registry.
• Anomaly detection models: Unsupervised and semi-supervised approaches for detecting novel emitters the system has never seen. Novelty detection, one-class classifiers, autoencoders, or hybrid approaches on RF spectral data. You complement the rule-based detection engine with learned representations that generalize.
• Production software platform: Containerized edge deployment (Docker on embedded Linux), CI/CD pipeline from git to edge device, OTA update infrastructure, fleet health monitoring, and the backend services — event ingestion, time-series storage, API layer — that make deployed sensor nodes observable and manageable. You use AI coding tools to accelerate infrastructure work and focus your time on the ML problems only you can solve.
• Data pipeline and dataset curation: Structured event output from edge nodes into the central data store. Collection protocols, labeling standards, data quality gates, augmentation pipelines, and dataset versioning. The quality of your training data determines the quality of your product.
• Model evaluation: Define metrics that matter operationally — probability of detection, false alarm rate, classification accuracy by threat type, performance under noise and interference. Your evaluation framework translates ML metrics into operator-relevant performance guarantees.

WHAT WE REQUIRE

3+ years of ML engineering with deployed production models. B.S. in Computer Science, Electrical Engineering, Mathematics, Physics, or related field required; M.S./Ph.D. preferred.

• Production ML: You have trained, validated, and deployed models that run in production — not research papers, not Kaggle competitions, not Jupyter notebooks. You understand the difference between training accuracy and operational performance.
• Signal/time-series classification: Experience with classification or anomaly detection on signal data, time-series data, or sensor data. RF domain experience preferred but teachable if ML fundamentals are strong and you learn fast.
• Deep learning frameworks: PyTorch or TensorFlow. You can build custom architectures, not just fine-tune pretrained models. CNN, RNN/LSTM, transformer, and autoencoder experience on non-image data.
• Edge deployment: TensorRT, ONNX Runtime, TFLite. Model quantization (INT8, FP16), pruning, and optimization for embedded inference. You’ve shipped models to hardware that isn’t a cloud GPU.
• Production software engineering: Strong Python and C/C++. Docker, CI/CD, REST/gRPC API development, Linux systems administration. You can build and deploy production backend services, not just train models. You treat software engineering as a core competency, not a chore.
• Data engineering: Feature pipelines, dataset management, versioning (DVC, MLflow, or equivalent). Time-series storage (PostgreSQL, TimescaleDB, or equivalent). You can build and maintain the data infrastructure your models depend on.
• Evaluation rigor: You design experiments properly. Cross-validation, stratified splits, confidence intervals, and operational metrics (Pd, Pfa, ROC/AUC). You don’t ship a model without understanding where it fails.
• Preferred: RF signal processing fundamentals — FFT, spectrogram interpretation, modulation basics. GNU Radio or SDR experience. SIGINT, EW, or radar ML experience. Experience with adversarial robustness or model performance under distribution shift. Active security clearance or ability to obtain.

WHAT WE OFFER

• Highly competitive base salary + founding equity.
• Act 60 Puerto Rico: Income tax on first $100K only, above that is exempt. 0% capital gains on equity for bona fide PR residents.
• Full health, dental, vision + flexible PTO.
• A founding role where you own the technical vision. Not a staff augmentation seat. Not someone else's roadmap.

This position requires access to export-controlled information under ITAR (22 CFR §120–130). Applicants must be U.S. citizens as defined by 8 U.S.C. §1324b(a)(3). iTERRA Solutions is an Equal Opportunity Employer. All qualified applicants will receive consideration without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, or veteran status.