Senior AI Framework Engineer

Modular•United States / Canada,

108d•$180,000 - $324,000•Hybrid

About The Position

At Modular, we are building a next generation AI platform to power modern applications and facilitate access to cutting-edge hardware. The MAX Framework is our developer-facing layer: it defines the APIs developers use to express models, integrate custom kernels, orchestrate execution, iterate on quality and ship systems into production. As an Senior AI Framework Engineer for MAX, you will own and evolve core APIs and developer experience for inference and training of AI models. You will work at the intersection of API design, systems engineering, and modern AI frameworks. Your output will be the specifications, abstractions, and reference implementations that make MAX feel coherent, powerful, and intuitive to use — while preserving performance, portability, maintainability and TCO.

Requirements

Significant experience designing and evolving developer-facing APIs (SDKs, frameworks, or platforms).
Strong understanding of modern AI frameworks and their design tradeoffs (e.g., PyTorch, JAX, TensorFlow, vLLM, XLA/MLIR-adjacent ecosystems).
Experience with inference and training systems (model execution graphs, compilation, runtime scheduling, distributed execution, checkpointing, performance tuning).
Fluency in one or more systems / performance languages (C++, Rust, Go) and one or more user-facing languages (Python; familiarity with Mojo is a plus).
Excellent taste for API ergonomics: naming, composability, types, error handling, configurability, and clarity.
Strong written communication: you can write specs that engineers can implement without ambiguity.
High engineering standards, pragmatism, and a bias towards incremental development without compromising long term design.

Nice To Haves

familiarity with Mojo is a plus

Responsibilities

Influence the design of the API surface for MAX: namespaces, core abstractions, extension points, programming model, compatibility guarantees and developer experience.
Design inference APIs that support real-world serving needs: model loading, distributed inference, quantization, tokenization/pipelines, configuration surfaces, batching/streaming, and deployment-oriented ergonomics.
Design training APIs that scale from single device to distributed execution, with clear primitives for device placement, parallelism, checkpointing, and observability.
Create a coherent programming model across Python and Mojo-adjacent surfaces: align naming, types, and conventions; avoid leaky abstractions; define the "pit of success".
Drive RFCs and technical specs: write and socialize proposals; gather feedback from internal model engineers and external users; iterate towards consensus.
Partner cross-functionally with compiler/runtime, kernels, cloud/serving, and documentation/DevRel teams to ensure APIs map cleanly to underlying capabilities.
Build reference implementations and exemplar code: golden-path examples, architecture templates, and best-practice patterns that teams can copy.
Set quality bars for APIs: versioning policy, deprecation strategy, test strategy, and documentation requirements.