Systems Architect

Oxman•New York City, NY

About The Position

OXMAN is seeking its first System Architect for EDEN, a generative ecosystem digital twin that aims to identify earthworks, planting, and management sequences that achieve environmental outcomes such as carbon sequestration, stormwater retention, microclimate cooling, and biodiversity. EDEN combines geospatial representation learning, physics-grounded world models, physical solvers, learned surrogate models, mathematical optimization, and AI agents to simulate ecosystem development and search for interventions whose outcomes can be predicted and validated. The System Architect will own technical architecture across EDEN’s platform, science, and applications teams. This hands-on role designs and builds the system that composes models, runs simulation and optimization loops at scale, supports trustworthy results, and delivers them through a unified interface. The architect will establish shared data and execution abstractions, create the runtime and search harness, bring research prototypes onto a portable foundation, embed validation in the platform, and establish engineering practices. The role is intended to help EDEN grow from research prototypes and one validated application into a cohesive system that can support many applications. During the first 90 days, the architect will map existing components and applications, document dependencies and end-to-end operation, write an architecture specification, and implement initial shared data contracts and a procedural graph runtime with input from the science and application teams. They will unify search and evaluation into a single cached lifecycle and take one application through the new architecture from user request to validated result.

Requirements

Substantial experience designing and owning complex compute- or data-intensive systems, ideally in simulation, scientific computing, production machine learning, mathematical optimization, or geospatial platforms.
A track record of turning research prototypes into reliable, well-structured products.
Strong systems design expertise and Python expertise.
Comfort with GPU workloads, parallel orchestration, and cloud infrastructure.
Care for numerical reproducibility and whether system outputs can be trusted, not only whether the system runs.
Clear writing and communication skills for scientists, designers, and engineers.
A hands-on approach, including personally building initial versions of critical system components.
Sound judgment about foundational decisions that are expensive to change later.
Comfort working across disciplines and teams, including platform engineering, scientific modeling, and applications.

Nice To Haves

Experience with black-box, Bayesian, or multi-objective optimization, quality-diversity search, or reinforcement learning.
Experience orchestrating and sandboxing LLM agents or program synthesis in production.
Experience with surrogate modeling, differentiable simulation, or GPU compute frameworks.
Experience with procedural, DAG, or node-based execution systems.
Experience integrating with 3D web visualization, CAD/BIM, or GIS software.
Interest in ecology, earth observation, or environmental modeling.

Responsibilities

Establish shared schemas for traits, physical fields, and world states, along with common contracts for operators and evaluators and a representation for EDEN’s procedural graphs.
Design the search, optimization, and agentic control loop, including translating user intent into formal problem bounds and establishing guardrails based on expert-validated physical and ecological limits.
Orchestrate parallel search, inner-loop tuning, evaluation budgets, and diagnostic feedback for AI agents and numerical optimizers.
Design and build a runtime for physical simulation, learned surrogate models, and long-running optimization at scale, including deterministic execution, content-aware caching, sandboxed code execution, and fidelity-versus-speed trade-offs.
Unify existing research and design prototypes and applications on a portable, scalable foundation that allows new scientific modules and application areas to integrate without custom glue.
Make validation a platform feature by ensuring models carry provenance, calibrated uncertainty, and documented conditions of validity, and by supporting determinism, out-of-sample testing, and physics-grounded checks.
Design geospatial and ecological data ingestion, versioned site and run state, backend services, APIs, and integrations that support EDEN’s web interface and external design tools.
Establish testing, reproducibility, continuous integration, and practices that distinguish research code from production code.
In the first 90 days, map components and applications, document their dependencies and end-to-end operation, produce an architecture specification, and implement initial shared data contracts and procedural graph runtime.
Unify agentic search, numerical optimization, and simulation through a single cached lifecycle, then take one application end-to-end through the new architecture from user request to validated result.