AI Accelerator Networking Architect (CCL)

Evollabs; Dubai; On-site; AED 250,000 - 450,000; Full time

Job summary

Evollabs in Dubai designs and develops AI accelerator hardware and software for datacenter-scale inference. This role focuses on designing and implementing collective communication for AI accelerators, including both NPU-native and heterogeneous NPU-GPU paths, at the intersection of runtime, networking, and hardware.

You will work on transport-layer integration across PCIe, Ethernet/RDMA, and custom fabrics, aiming to optimize latency, bandwidth, and throughput for distributed AI workloads in

Qualifications

• 5+ years of relevant experience with MS/PhD or equivalent in CS/CE.
• Strong C/C++ and Python programming skills with low-level runtime, driver, networking, or distributed systems experience.
• Deep understanding of collective communication primitives (AllReduce, ReduceScatter, AllGather, AllToAll, Broadcast, Send/Recv, Barrier).
• Familiarity with NCCL, RCCL, oneCCL, MPI, UCX, or libfabric.
• Knowledge of GPU/NPu memory models and heterogeneous device communication.
• Experience with accelerator runtimes: queues, streams, events, submission, synchronization.
• Understanding hardware-software interactions in accelerator systems (DMA, doorbells, fences).

Responsibilities

• Design and implement collective communication primitives for AI accelerators.
• Build NPU-native CCL support for multi-chip and rack-scale AI inference systems.
• Design and implement heterogeneous NPU-GPU communication paths for distributed AI workloads.
• Develop communication runtime components for device discovery, rank management, topology awareness, stream/queue integration, events, and synchronization.
• Work on transport-layer integration across PCIe, Ethernet/RDMA, custom accelerator fabrics, or other device-to-device communication mechanisms.
• Collaborate with hardware, runtime, compiler, and model-serving teams to co-design scalable communication paths for AI workloads.
• Profile, debug, and optimize communication bottlenecks including bandwidth, latency, congestion, synchronization overhead, and compute-communication overlap.
• Build correctness, performance, and stress tests for multi-device and heterogeneous communication scenarios.

Skills

C/C++

Python

Distributed systems

NCCL/RCCL

PCIe/Networking

GPU/NPu memory

Runtime development

Performance optimization

Education

MS/PhD CS/CE or equivalent

Tools

MPI

UCX

libfabric

NCCL

OneCCL

Job description

Evollabs in Dubai designs and develops AI accelerator hardware and software for datacenter-scale inference. This role focuses on designing and implementing collective communication for AI accelerators, including both NPU-native and heterogeneous NPU-GPU paths, at the intersection of runtime, networking, and hardware.

You will work on transport-layer integration across PCIe, Ethernet/RDMA, and custom fabrics, aiming to optimize latency, bandwidth, and throughput for distributed AI workloads in