AI Frameworks Engineer

Great Sky•Boulder, CO

2d•$160,000 - $220,000•Hybrid

About The Position

Great Sky is a technology startup based with a mission to rebuild AI from first principles. We are pursuing neuroscience-inspired hardware and algorithms that overcome the greatest challenges in scaling AI systems. We are looking for an AI Frameworks Engineer to help build the core software layer that connects our research ideas to working, testable, scalable AI systems. This is a software engineering role focused on the internal AI framework: the abstractions, APIs, graph execution paths, training strategies, validation systems, and test infrastructure that make our research stack usable and reliable.

Requirements

Bachelor's or Master's degree in Computer Science or a related field, or equivalent professional experience.
Strong Python software engineering skills, with experience building libraries, frameworks, or infrastructure used by other engineers or researchers.
Hands-on experience with at least one modern ML framework, especially JAX or PyTorch.
Proficiency with training-loop internals: forward passes, losses, gradients, optimizers, metrics, parameter/state handling, and numerical testing.
Experience designing clean APIs and abstractions in complex codebases.
Ability to reason about graph-structured computation, model execution, dependency boundaries, and runtime state.
Strong testing instincts: unit tests, integration tests, regression tests, numerical parity tests, and clear failure modes.
Ability to work in a research-heavy environment where requirements evolve and the right abstraction often emerges through iteration.
Excellent written and verbal communication skills, especially when translating between research ideas and maintainable engineering designs.

Nice To Haves

Prior experience working in an AI technology environment.
Experience with typed Python, protocol-oriented design, dataclasses, Pydantic, package refactors, or public/internal SDK design.
Familiarity with multi node and multi gpu training for JAX or PyTorch
Experience with custom kernel writing in Triton/Cuda/Pallas
Experience designing plugin or registry systems for losses, optimizers, training strategies, model components, or execution backends.
Experience with compiler-adjacent or graph-adjacent systems: computational graphs, schedulers, IRs, topology validation, graph transformations, or hardware-aware execution.
Experience working near hardware teams, accelerator teams, simulation teams, or low-level performance work.
Familiarity with containers, cloud execution, job orchestration, experiment tracking, or production training harnesses.

Responsibilities

Design, implement, and maintain core AI framework abstractions for defining, training, validating, and executing novel graph-structured models.
Build clean Python APIs for training workflows, including loss objects, model runtime bundles, optimizer/training strategy interfaces, and more.
Develop and maintain JAX and PyTorch infrastructure for model execution, customizable training protocols, numerical parity tests, and reusable forward-pass utilities.
Help design validation systems that catch incompatible model/loss/solver/topology combinations early, with clear diagnostics for researchers and production users.
Organize research code into stable, well-tested framework components without slowing down exploration.
Create clean seams between research-mode workflows and production harness workflows, so the same infrastructure can support ad hoc experiments, custom training loops, and ticketed/cloud training runs.
Write high-quality tests for numerical behavior, protocol conformance, registry behavior, graph execution and training loops.
Collaborate closely with researchers and hardware/software engineers to translate evolving research requirements into robust software architecture.
Improve developer ergonomics: documentation, examples, error messages, package organization, type boundaries, and API consistency.
Over time, help shape the broader AI software architecture that connects model design, simulation, training, validation, and hardware execution.