Principal Engineer, On-Device Inference, GPU Kernels

Opera

. Train, tune, and ship draft models for speculative decoding, including block-diffusion drafters, semi-autoregressive drafters, EAGLE-style heads, and multi-token prediction .

Posted 11/9/2026full-timeWrocław • PolandLeadWebsite

Core Competencies

Role fit

Core Competencies

Use this summary to align your resume positioning with the role.

Demonstrates expertise in GPU programming with a strong focus on CUDA, Metal/MLX, and Vulkan, alongside hands-on experience in speculative decoding and quantisation techniques. Proven ability to enhance model performance through innovative caching strategies and rigorous validation of numerical correctness.

Highest-signal resume keywords

GPU ProgrammingSpeculative DecodingQuantisation TechniquesPerformance OptimizationNumerical Correctness

ATS Keywords

Tailor your resume

Applicant Tracking System Keywords

Tip: use these terms in your resume and cover letter to boost ATS matches.

Hard Skills

CUDAMetal/MLXVulkanSpeculative DecodingQuantisationPerformance TelemetryHierarchical KV CachingAttention MechanismsMixed-Precision SchemesKernel-Level Scheduling

Tools & Technologies

OpenVINOLlama.cppGgmlMLXVLLMSGLangONNX RuntimeGPTQAWQMX Formats

Industry Keywords

AI CapabilitiesInference WorkloadsLatencyMemory OptimizationConsumer Hardware

Tech Stack

Tools & technologies

Flash

About the role

Key responsibilities & impact

Train, tune, and ship draft models for speculative decoding, including block-diffusion drafters, semi-autoregressive drafters, EAGLE-style heads, and multi-token prediction
Write, fuse, and tune kernels for attention, quantised matmul, MoE routing, sampling, and draft verification across Metal/MLX, CUDA, Vulkan, and OpenVINO
Design mixed-precision schemes for weights, activations, and KV cache across consumer hardware
Improve time-to-first-token and tokens per second through flash and paged attention, sliding-window and hybrid attention, chunked prefill, and kernel-level scheduling
Build hierarchical KV and context caching across GPU memory, system RAM, and SSD, with prefix reuse, eviction, and prefetching
Build test matrices, performance telemetry, and regression gates for drivers, numerical correctness, and UI performance
Add runtime instrumentation hooks for activations and attention
Work closely with Opera's AI leadership and contribute to on-device AI capabilities used by millions of Opera users

Requirements

What you’ll need

Deep, hands-on GPU programming in at least one of CUDA, Metal/MLX or Vulkan compute, plus evidence of carrying that work over to a second backend
Hands-on speculative decoding experience, including implementing or training drafters and understanding acceptance rate and end-to-end speed-up
Track record of making real inference workloads measurably faster in tokens per second, latency or memory, with explainable numbers
Strong command of quantisation and its effect on model quality, including how quality is measured
Discipline about correctness, including validating numerics against a reference and measuring quality
Contributions to llama.cpp/ggml, MLX, vLLM, SGLang, ONNX Runtime, OpenVINO or similar are nice to have
Experience with drafter-training frameworks or published work on speculative or parallel decoding is nice to have
Quantisation research or tooling experience with GPTQ, AWQ, k-quants, MX formats, or quantisation-aware fine-tuning is nice to have
KV cache compression, offloading or paging at the systems level is nice to have
Experience with NPUs such as Intel, Qualcomm, AMD, or Apple Neural Engine is nice to have
CV must be submitted in English

Benefits

Comp & perks

Access to Apple Silicon, NVIDIA (including Blackwell-class), AMD, and Intel hardware, plus cloud GPUs for model training
Opportunities for professional growth
Technical ownership of performance-critical development
Opportunities to contribute to open-source projects and publish technical findings where appropriate
Collaborative, international team with a flat structure and direct communication
Inclusive environment with opportunities to grow and make an impact