AI Inference & Compression Engineer

Beijing Foreign Enterprise Management Consultants Co.,Ltd.; Singapore; On-site; SGD 180,000 - 300,000; Full time

Job summary

Huawei in Singapore seeks an AI Inference & Compression Engineer to push the boundaries of LLM serving and video coding.

You will research compression algorithms, optimize KV caches and model quantization, and tackle memory bandwidth bottlenecks during autoregressive decoding.

In addition, you’ll develop AI-based video coding components, bridge R&D with production, and assess performance with PSNR, VMAF, latency, and throughput metrics.

Qualifications

• Master’s or PhD in Computer Science, Electronic Engineering, Mathematics, or related fields (PhD preferred).
• Solid understanding of video coding fundamentals including prediction, transform coding, quantization, and entropy coding with hands-on experience in standards such as H.265/HEVC, AV1, or H.266/VVC.
• Strong understanding of Transformer architectures and attention mechanisms, as well as key performance bottlenecks in generative AI inference, particularly memory bandwidth constraints (“memory wall”).
• Strong proficiency in Python and C/C++. Hands-on experience building, training, and modifying models using PyTorch, TensorFlow, etc.

Responsibilities

• LLM Inference Acceleration. Research and develop advanced compression algorithms to accelerate LLM serving. Focus on KV cache optimization, model quantization, and resolving memory bandwidth bottlenecks during autoregressive decoding.
• Classical Codec Development. Design and implement advanced video compression algorithms, focusing on improving RD performance, optimizing entropy coding, and enhancing quantization design for real-world applications.
• AI-Based Media Coding. Develop and optimize AI-based video coding components, including AI-based loop filters, optical flow, and intelligent rate control.
• Model Deployment & Fusion. Bridge the gap between AI research and production. Optimize deep learning models for efficient inference and ensure seamless integration of compression algorithms into deployment frameworks (e.g., vLLM).
• Performance & Quality Evaluation. Conduct rigorous objective and subjective visual quality assessments such as PSNR and VMAF for video systems, as well as perplexity, zero-shot benchmarks, latency, and throughput analysis for LLM systems.

Skills

Python

C/C++

PyTorch

TensorFlow

Transformer architectures

Memory bandwidth

Education

Master’s or PhD in Computer Science/EE/Math

Tools

CUDA

HEVC/AV1/H.266 codecs

Job description

On behalf of Huawei, a world-renowned information and communication technology company, we are seeking passionate and talented individuals to join our team as AI Inference & Compression Engineer.

Key Responsibilities

• LLM Inference Acceleration. Research and develop advanced compression algorithms to accelerate LLM serving. Focus on KV cache optimization, model quantization, and resolving memory bandwidth bottlenecks during autoregressive decoding.
• Classical Codec Development. Design and implement advanced video compression algorithms, focusing on improving Rate–Distortion (RD) performance, optimizing entropy coding, and enhancing quantization design for real-world applications.
• AI-Based Media Coding. Develop and optimize AI-based video coding components, including AI-based loop filters, optical flow, and intelligent rate control.
• Model Deployment & Fusion. Bridge the gap between AI research and production. Optimize deep learning models for efficient inference and ensure seamless integration of compression algorithms into deployment frameworks (e.g., vLLM).
• Performance & Quality Evaluation. Conduct rigorous objective and subjective visual quality assessments such as PSNR and VMAF for video systems, as well as perplexity, zero-shot benchmarks, latency, and throughput analysis for LLM systems.

Required Qualifications

• Master’s or PhD in Computer Science, Electronic Engineering, Mathematics, or related fields (PhD preferred).
• Solid understanding of video coding fundamentals including prediction, transform coding, quantization, and entropy coding with hands‑on experience in standards such as H.265/HEVC, AV1, or H.266/VVC.
• Strong understanding of Transformer architectures and attention mechanisms, as well as key performance bottlenecks in generative AI inference, particularly memory bandwidth constraints (“memory wall”).
• Strong proficiency in Python and C/C++. Hands‑on experience building, training, and modifying models using PyTorch, TensorFlow, etc.

Preferred Qualifications

• ISP Knowledge. Familiarity with Image Signal Processing flow, such as demosaicing, denoising, and tone mapping.
• Image Processing. Experience in computer vision-based image enhancement (e.g., de‑blurring, artifact removal, or HDR).
• Hardware Optimization. Knowledge of SIMD, CUDA, or other hardware acceleration techniques for video and tensor processing.