Data Science Engineer

Digantara•Denver, CO

3d•Onsite

About The Position

Digantara U.S. is a leading Space Surveillance and Intelligence company focused on ensuring orbital safety and sustainability. With expertise in space-based detection, tracking, identification, and monitoring, Digantara provides comprehensive domain awareness across regimes, allowing end users to have actionable intelligence on a single platform. At the core of its infrastructure lies a sophisticated integration of hardware and software capabilities aligned with the key principles of situational awareness: perception (data collection), comprehension (data processing), and prediction (analytics). This holistic approach empowers Digantara to monitor all Resident Space Objects (RSOs) in orbit, fostering comprehensive domain awareness. Digantara U.S. is seeking an experienced and driven Data Science Engineer to develop the machine-learning components and integration infrastructure supporting the company's SDA analytics pipelines. Across programs, the role builds trajectory-classification and anomaly-detection models that operate on orbit-determination output, and maintains the benchmarking and evaluation frameworks that establish whether those pipelines perform under sparse, gapped, and noisy observation conditions. Emphasis is placed on the characteristics that determine operational value: false-positive behavior under degraded observations, calibration of confidence metrics suitable for operator use, and inference cost compatible with constrained onboard processing. The role works closely with astrodynamics and embedded-systems staff to ensure models reflect genuine dynamical structure and remain deployable within onboard resource limits.

Requirements

BS/MS in Computer Science, Applied Mathematics, Statistics, Aerospace Engineer, Physics, or a related quantitative field, plus [4]+ years of applied ML experience (or equivalent).
Strong Python and the scientific stack (NumPy, SciPy, pandas); fluency in at least one deep-learning framework (PyTorch preferred).
Demonstrated experience building ML systems on time-series, sequential, or state-estimation-adjacent data, rather than tabular or vision benchmarks alone.
Sound evaluation methodology: class imbalance, calibration, uncertainty quantification, and the failure modes of small or synthetically generated datasets.
Software-engineering discipline sufficient for a shared codebase: version control, testing, reproducible environments, and documented interfaces.
Clear technical writing for customer-facing deliverables.
Must be able to obtain and hold a U.S. security clearance

Nice To Haves

Experience with physics-informed ML, or hybrid approaches that embed dynamical structure into learned models.
Familiarity with orbit determination, tracking, or multi-target data association (JPDA, MHT, or similar).
Experience with model compression, quantization, or deployment to constrained and embedded targets.
Prior work on government R&D programs (SBIR/STTR, AFRL, DARPA, Space Force) and familiarity with TRL terminology.
Experience with anomaly detection where anomalies are rare, poorly labeled, or defined only by a physical model.

Responsibilities

Develop AI/ML models for trajectory classification across orbit regimes and families, and for detection of anomalous dynamical behavior.
Integrate and maintain end-to-end analytical pipelines spanning observation processing, hypothesis generation, orbit estimation, propagation, and classification; define interfaces, own the shared, reproducible codebase.
Build benchmarking and evaluation frameworks: estimator convergence behavior, classification accuracy and confusion structure, false-positive and false-negative characterization, time-to-custody, and sensitivity to track gaps and elevated measurement uncertainty.
Design experiments that distinguish genuine generalization from dataset artifacts held-out families, degraded-observation ablations, and cross-checks against independent reference datasets.
Produce calibrated confidence metrics suitable for downstream operational use, documented precisely enough to support operator decisions.
Partner with embedded-systems staff to characterize model complexity, memory footprint, and inference latency, and to identify quantization, pruning, or architectural simplifications that perform within deployment constraints.
Contribute machine-learning inputs to CONOPS and systems-engineering activities, including data-flow definition, model lifecycle and retraining considerations, and critical technology element identification.