Applied AI/ML Engineer

Ursa Space Systems•Ithaca, NY

4d•$185,000 - $199,000•Remote

About The Position

Ursa Space turns complex satellite and spatial data into decision-ready answers. Our agentic GeoAI platform lets users ask a question in plain English about a place, an event, or an activity anywhere on Earth and handles the rest: selecting the right data sources, tasking sensors, running analytics, and delivering an insight report in minutes instead of hours. We're looking for an Applied AI/ML Engineer to help build the intelligence behind that platform. This is a hybrid role by design. Some weeks you'll be deep in computer vision training and deploying object detection and segmentation models on SAR and electro-optical satellite imagery. Other weeks you'll be building agentic systems: designing tool-calling workflows, orchestrating LLM-driven analysis pipelines, and building the evaluation infrastructure that keeps them reliable. The work varies significantly project to project, and the right candidate sees that as a feature, not a bug. You'll report to the Director of Analytics and work side-by-side with data scientists, image scientists, product owners, and customers. This role may include pre-scheduled on-call rotations requiring occasional evening or weekend technical support. This is a virtual role and an exempt position.

Requirements

B.S. or M.S. in Computer Science, Data Science, or a related STEM field
4–6 years of experience building and deploying machine learning systems for product- or software-focused organizations
Strong Python and production software engineering practices: Git, Docker, testing, code review, CI/CD
Experience training and deploying deep learning models for computer vision tasks (object detection, image segmentation) using PyTorch or similar frameworks
Hands-on experience building LLM-powered applications: prompt design, structured outputs, tool use / function calling, and agentic architectures
Experience evaluating ML systems with appropriate metrics — both traditional (RMSE, FPR/TPR, AUC/ROC, IoU) and LLM/agent evaluation approaches
Experience developing and deploying in AWS environments
Strong communication skills: able to present findings and explain complex systems to technical and non-technical audiences, including customers

Nice To Haves

Prior experience with remote sensing data especially synthetic aperture radar (SAR), GIS, and/or spatial statistics
Experience fine-tuning foundation models or VLMs (e.g., LoRA/PEFT, multimodal adaptation for domain-specific imagery)
Familiarity with agent orchestration frameworks and protocols (e.g., LangGraph, Claude Agent SDK, MCP) and LLM observability/eval tooling
Geospatial Python stack: GDAL, rasterio, geopandas, xarray
Experience with SQL/NoSQL databases and vector stores
Experience leading projects on complex, cross-discipline teams
Familiarity with Confluence, Jira, and Miro
Regression and drift detection: did a "small" model or prompt change silently degrade accuracy on a case type that isn't in the dev team's day-to-day test set?
Adversarial and edge-case testing deliberately probing for failure modes (ambiguous imagery, conflicting sources, out-of-distribution inputs) rather than testing the happy path a developer already validated.
Confidence calibration to ensure that a system's stated confidence actually corresponds to real-world correctness rates, which is its own measurement discipline.
Traceability and reproducibility to demonstrate why the system produced a given output, on a specific input, at a specific model/data version.

Responsibilities

Design, build, and maintain agentic AI systems that automate stages of the geospatial analysis cycle from natural-language question intake through data selection, multi-source analysis, and report generation
Develop tool-use, orchestration, and context-management capabilities that connect LLMs to our geospatial data services, analytics, and 90+ integrated data feeds
Conduct Verification and Validation (V&V) against AI/ML and LLM outputs to ensure accurate resulting information
Train, fine-tune, evaluate, and deploy computer vision models (object detection, segmentation, change detection) on SAR, EO, and other Earth Observation data
Adapt and fine-tune vision-language models (VLMs) to move beyond bounding boxes toward full scene understanding and extracting the meaningful content of imagery, not just locating objects
Build evaluation and observability infrastructure for both classical ML (precision/recall, IoU, AUC/ROC) and LLM/agent behavior (task success, groundedness, regression testing), and use it to drive measurable improvement
Own projects end to end: from data definition and prototyping through production deployment, validation, and maintenance
Work directly with product owners, internal platform users, and external customers to understand needs, translate requirements into agentic system designs, and explain how the technology works to people who experience it as a black box
Integrate third-party and multi-source data sets into analysis pipelines
Do ad-hoc analysis and answer time-sensitive questions from stakeholders across the organization
Act as a technical resource for teammates to bring awareness of new models, techniques, and tools that help the whole team grow
Owning measuring actual observed error against budget, not just believing the budget on paper.
Adversarial and edge-case testing deliberately probing for failure modes (ambiguous imagery, conflicting sources, out-of-distribution inputs) rather than testing the happy path a developer already validated.
Confidence calibration to ensure that a system's stated confidence actually corresponds to real-world correctness rates, which is its own measurement discipline.
Traceability and reproducibility to demonstrate why the system produced a given output, on a specific input, at a specific model/data version.
All other duties as assigned.