The Opportunity:
At Booz Allen, we power possibility for our clients facing their toughest challenges. Our cross-domain team of consultants, technologists, researchers, engineers, and cyber experts builds innovative solutions for the infrastructure and capabilities of tomorrow. We're trusted partners on our clients' most critical missions, and we bring expertise to the table, but it's our attitude, our courage, curiosity, and passion for tomorrow, that makes us unique.
We're seeking an AI/ML Engineer to join a team building a converged platform that runs real-time wireless signal processing and AI inference on the same GPU-accelerated hardware. You will develop, optimize, and deploy models that run alongside live network workloads, where latency budgets are measured in microseconds and compute is shared with the radio stack. The work spans the full lifecycle: data pipelines from network telemetry, model training and benchmarking, GPU-optimized inference, and containerized deployment on Kubernetes in a hands-on lab.
What You Will Do:
- Design, train, and evaluate ML models for wireless network use cases, including traffic prediction, anomaly detection, interference classification, resource scheduling, and energy optimization, using network telemetry, KPIs, and RF data captured in the lab
- Build data pipelines that ingest, label, and version telemetry from the network stack, such as performance counters, logs, packet captures, and I/Q samples, into reproducible training datasets
- Optimize trained models for real-time inference on GPU-accelerated edge servers: quantization, pruning, TensorRT or ONNX Runtime conversion, CUDA profiling, and latency and throughput benchmarking against hard targets
- Package and deploy models as containerized inference services on Kubernetes using NVIDIA Triton Inference Server, the NVIDIA GPU Operator, and MIG or time-sliced GPU sharing alongside live network workloads
- Develop AI-driven network applications such as xApps and rApps that consume network telemetry and return control decisions through standardized interfaces, and validate them end to end in the lab
- Instrument the platform with MLOps tooling: experiment tracking, model registry, automated retraining triggers, drift monitoring, and CI/CD for model deployment
- Benchmark GPU resource contention between signal processing and AI inference workloads, and characterize and document the operating envelope for co-location on a single platform
- Apply security and assurance practices to the ML lifecycle: model provenance, dataset lineage, adversarial robustness testing, and alignment with DoD AI assurance guidance
- Write and maintain technical documentation, including model cards, training and deployment procedures, benchmark reports, and reproducibility guides, concurrently with lab work, not after the fact