Open to opportunities

Xuemei Jiang

@xuemeijiang

Message

I build production ML systems for large-scale image retrieval, visual inspection, and AI agents.

Germany

Message

What I'm looking for

I'm looking to build production ML systems involving embedding models, retrieval at scale, computer vision, and AI-agent backends.

I've built production machine learning systems at Everest Systems and the United Nations’ WIPO, spanning AI-agent backends, large-scale image retrieval, and visual defect inspection.

At WIPO, I trained embedding models on 22 million images using multi-GPU DDP, improving top-5 accuracy by 10% while cutting training time by 30%. I designed multi-view retrieval with late fusion, deployed Milvus across 10M+ documents for cross-lingual retrieval, and created evaluation frameworks for recall, hit rate, and latency.

I also ship practical ML infrastructure: self-hosted VLMs for compliant on-premise patent data, Dockerized FastAPI services, and testing pipelines for multi-agent coding systems. My research on self-supervised learning in remote sensing has been presented orally at IGARSS.

Experience

Work history, roles, and key accomplishments

AI Engineer

Everest Systems

Dec 2025 - May 2026 (5 months)

Rebuilt the reporting agent's data model to isolate sandbox development from production during migrations. Owned testing pipeline for spec-to-app coding agent using multi-agent architecture, catching ~10% runtime errors per generation.

Python TypeScript Docker fastAPI AWS CI CD MLFlow W&B Git Linux Multi Agent Systems Testing

Machine Learning Engineer

United Nations (WIPO)

Aug 2024 - Nov 2025 (1 year 3 months)

Trained embedding models on 22M images using multi-GPU DDP, achieving +10% top-5 accuracy and -30% training time. Designed multi-view retrieval with late fusion, improving recall by 3.5% over DINOv2 ViT baseline, and deployed self-hosted VLMs via vLLM for compliance.

Python PyTorch NumPy Computer Vision Distributed Training Embeddings Transformers Vector Search Milvus RAG VLLM Evaluation Benchmarking Docker fastAPI AWS CI CD MLFlow W&B Git Linux

Machine Learning Engineer Intern

United Nations (WIPO)

Feb 2024 - Aug 2024 (6 months)

Deployed Milvus vector search over 10M+ documents for cross-lingual retrieval, increasing recall@10 to over 60%. Built a golden QA dataset with 3,000+ labeled patent examples for RAG evaluation.