The NPU SDK part of Software Team owns the end-to-end SDK user experience: modeling work to bring state-of-the-art AI models (LLM, Multi-modal, Vision, Speech, etc.) onto Rebellions NPU, building and operating CI/CD pipelines to ship a production-ready SDK to external users, releasing and distributing the software, and authoring and publishing developer documentation.

Responsibilities and Opportunities

• SDK Ecosystem & Release Pipeline: Design, build, and operate CI/CD pipelines that automate and scale the SDK release process
• Model Deployment & Porting: Port and optimize the latest deep learning models (LLM, Multi-modal, etc.) onto Rebellions NPU, analyze performance bottlenecks, and validate numerical parity against GPU reference implementations when needed
• Framework Integration: Integrate the SDK tightly with the open-source inference and serving ecosystem, including Hugging Face Transformers, Diffusers, vLLM, and to maximize the developer experience
• Verification & Debugging: Design model-feeding frameworks and debugging utilities to ensure functional correctness and long-term stability of the SDK
• Developer Documentation: Author and maintain developer-facing documentation — API references, tutorials, model support matrices, and release notes — and operate the documentation build and publishing pipeline for external developers and partners

Key Qualifications

• Master’s degree or above in Computer Science, Electrical Engineering, or a related field
• Deep working knowledge of CI/CD tools such as GitHub Actions, Airflow, and Buildkite
• Deep understanding of modern deep learning model architectures, including LLMs (Transformer family), Multi-modal, Vision, and Speech
• Strong familiarity with PyTorch internals, including model customization and graph transformations
• High-performance software development skills in Python and Modern C++ (C++17 or later)
• Experience using and integrating major open-source libraries such as Hugging Face (Transformers, Diffusers) and vLLM
• Hands-on experience with Python and Kubernetes-based environments for deploying and troubleshooting AI inference workloads in production

Ideal Qualifications

• 5+ years of relevant industry experience, or equivalent practical expertise
• Experience optimizing and deploying models on specific hardware using CUDA, TensorRT, TensorRT-LLM, MLIR, Triton, or TVM