Job Summary

We are hiring acore algorithm R&D engineerto develop and advance the key AI capabilities of ourinternally developed vision platform. You will drive research-to-production delivery ofstate-of-the-artcomputer vision, deep learning, and multimodal foundation model techniques, focusing on industrial-grade performance, robustness, and efficiency.

Key Responsibilities

Core Vision Algorithm R&D (Deep Learning + Transformers)

• Research, develop, and optimize computer vision algorithms across:
  CNN-basedclassification, anomaly detection, Siamese networks, object detection,rotated object detection, semantic segmentation, instance segmentation,keypointdetection.

• Build and improveTransformer-baseddetection/recognition architectures and training pipelines.

• Design evaluation protocols, run ablation studies, and iterate based on measurable improvements (accuracy, robustness, latency).

Few-shot / Small-sample Learning for Industrial Use Cases

• Own R&D forfew-shotrotated detection, segmentation, and anomaly detection—aiming to train effective models fromonly a few images.

• Explore and implement methods such as meta-learning, prompt-/prototype-based learning, retrieval-enhanced approaches, and foundation-model feature adaptation for industrial inspection scenarios.

LLM / VLM Fine-tuning & Reinforcement Learning (Post-training)

• Understand LLM/VLM principles and implement practical post-training pipelines: 

• Supervised fine-tuning(SFT), parameter-efficient fine-tuning (e.g.,LoRA/PEFT), alignment methods (e.g., RLHF/DPO-like approaches), evaluationharnessesand safety/quality checks.

• Build reproducible training workflows (data curation, experiment tracking, model versioning, deployment readiness).

Vector / Graph-based Learning for CAD/PCB & Structured Data

• Research and develop models beyond raster images forvector datascenarios (e.g., engineering drawings, PCB schematics/layouts), aiming to outperform image-based baselines.

• Applygraph neural networks(GNNs)and vector/geometric representations to tasks such as component understanding, connectivity reasoning, and structured recognition.

High-performance Implementation &Productionization

• Write efficient, maintainable code inC++ and Pythonfor training/inference pipelines and algorithm modules.

• Develop high-performancecomputekernels and optimizations usingSIMDand/orCUDA, profiling and improving runtime, memory use, and throughput.

• Collaborate with platform/software teams to integrate algorithms into product modules and ensure test coverage, stability, and maintainability.

Paper Reading & Reproducibility

• Regularly read and analyze top-tier papers;identifykey contributions and reproduce core algorithms in code.

• Deliver internal technical notes and share learnings with the team.

Required Qualifications

• Bachelor’s / Master’s / PhD in Computer Science, Electrical Engineering, Applied Mathematics, or related fields (industry experience may substitute).

• Strong fundamentals and hands-on experience indeep learning for computer vision, including detection and segmentation.

• Solid engineering ability withPython + C++; capable of building clean training code（withPytorch)and production-ready modules. 

• Practical experience with performance optimization and acceleration (one or more ofCUDA / SIMD / parallel computing).

• Ability to communicate effectively inboth Chinese(Mandarin)and English as the successful person will have to liaise with our counterparts in China.