Senior Staff / Principal Machine Learning Scientist, AI Inference & Optimization

Netskope•Santa Clara, CA

About The Position

About Netskope Today, there's more data and users outside the enterprise than inside, causing the network perimeter as we know it to dissolve. We realized a new perimeter was needed, one that is built in the cloud and follows and protects data wherever it goes, so we started Netskope to redefine Cloud, Network and Data Security. Since 2012, we have built the market-leading cloud security company and an award-winning culture powered by hundreds of employees spread across offices in Santa Clara, St. Louis, Bangalore, London, Paris, Melbourne, Taipei, and Tokyo. Our core values are openness, honesty, and transparency, and we purposely developed our open desk layouts and large meeting spaces to support and promote partnerships, collaboration, and teamwork. From catered lunches and office celebrations to employee recognition events and social professional groups such as the Awesome Women of Netskope (AWON), we strive to keep work fun, supportive and interactive. Visit us at Netskope Careers. Please follow us on LinkedIn and Twitter @Netskope . Positions are available at Senior Staff and above. Candidates are assessed individually and leveled according to their specific skills and background. About the role As a Senior Staff Machine Learning Scientist, you own the inference and optimization layer that makes AI in agentic workflows fast, efficient, and production-grade. You fine-tune and evaluate models, push latency and throughput on real hardware, and build the runtime that executes bounded AI tasks, validated against usage from Netskope’s large customer base so you optimize where the data points, not where you guess.

Requirements

10+ years of overall industry experience , with 4+ years hands-on in ML/AI (model development, fine-tuning, and inference optimization).
Hands-on with fine-tuning (e.g. LoRA/QLoRA), quantization (GGUF/AWQ/GPTQ), and inference runtimes (vLLM/SGLang, TensorRT-LLM, ONNX Runtime, llama.cpp, or MLX/CoreML).
Strong Python; comfort reaching into C++ for low-level interop is a plus.
Solid grasp of transformer internals and the levers that move real inference performance and cost: KV cache, attention, batching, memory footprint.
Fluency with agentic coding systems and genuine curiosity about agent harnesses like Claude Code, Pi, and Codex , so you should already be building with them, or itching to.
Clear communication: able to distill a model or infra bottleneck into an actionable concept for cross-functional teammates.

Nice To Haves

On-device or edge inference experience is a strong plus.

Responsibilities

Build and optimize the model inference path : quantization, KV-cache optimization, batching, and latency/memory/throughput tuning on constrained, commodity hardware.
Fine-tune and evaluate models for bounded tasks; build eval harnesses that gate a capability to release on real accuracy, latency, and security relevance.
Design and grow the task execution runtime (bounded sub-agents), pushing toward dynamic task generation and context compaction.
Drive hardware acceleration / sparsity and support for larger models as the platform matures.
Partner with the systems and backend engineers to ship capabilities end-to-end and iterate on real production signals.