Zhongyuan Wang
AI Scientist
Specializing in Agentic AI, LLM systems, and ML. Patent applications and enterprise PoC conversions.
Experience
Founding ML Scientist · RaptorX.AI, Remote2024 - Jul 2026
- Built a production-style AI engineering workflow for bank fraud alert investigation, covering orchestration, guardrails, case management, evidence/report generation, model training, evaluation, and MLOps.
- Built criteria-driven multi-agent systems for paper/patent production; owned technical direction, experiment validation, drafting, counsel handoff, and auditability.
- Designed and implemented multiple GCN-based models to improve the analysis and identification of fraud risks in financial networks.
- Converted PoCs into signed enterprise customers.
Senior Engineer, AI & Data Platform for Autonomous Driving · Lotus Tech (Geely Group), Hangzhou2021 - 2024
- Led an in-house video anonymization pipeline from vehicle data ingestion through data-center processing for fleet-scale autonomous-driving data.
- Trained and deployed face/license-plate detection across large-scale video datasets, achieving high recall for privacy-sensitive objects.
- Built a Kubernetes-based GPU scheduling platform for PyTorch workloads on multi-GPU infrastructure, improving utilization and reducing cloud costs.
- Optimized GPU-accelerated batch inference for fleet-scale video processing, reducing latency and improving throughput.
- Replaced third-party vendors with an internal ML platform, reducing costs while supporting GDPR/PIPL compliance.
Independent ML Study & Graph-ML Projects2020 - 2021
- Self-directed study of graph machine learning: implemented GNN models for node classification and link prediction on public datasets.
Software Development Engineer, Financial Anti-Fraud · DataVisor, Shanghai2020
- Built an unsupervised loan application fraud model combining GNN and traditional ML across behavioral, device, and transaction signals.
- Contributed to enterprise contract wins through technical leadership and client demonstrations.
Research Intern, DiDi AI Labs · UCL-DiDi Master Research Program, Beijing2019
- Leveraging Graph Neural Networks for user travel intent prediction, supervised by Prof. Jieping Ye (VP of DiDi Chuxing, IEEE Fellow).
- Selected as 1 of only 2 projects (out of 30+ candidates) across the entire UCL cohort for the DiDi AI Labs research program.
Publications
- When the Tool Decides: LLM Agents Defer Blindly to GNN Tools
First author, under review (TMLR) · arXiv:2606.14476
- LLM Features Can Hurt GNNs: Concatenation Interference on Homophilous Graph Benchmarks
First author, under review (TMLR) · arXiv:2606.17579
- How Much Do Deep GNNs Actually Help? Decomposing Depth, Features, and Structure
First author, under review (NeurIPS 2026 E&D)
Patents
Inventor on patent applications spanning graph-based fraud detection and LLM/agent systems.
Technical Skills
Agent / ML
Agentic engineering, LLM APIs, function calling/tool use, guardrails, agent evaluation, PyTorch, PyTorch Geometric (PyG), GNNs, autoencoders, fraud-ring detection
Data & MLOps
Spark / PySpark, Kafka, ClickHouse, Redis, Parquet, Docker, Kubernetes, ArgoCD, CI/CD, FastAPI / gRPC, MLflow
Cloud
AWS (S3 / EC2 / ECR), cloud GPU workflows, Linux, Grafana, Prometheus, Loki
Education
MSc, Data Science and Machine Learning · University College London2018 - 2019
- Thesis: Graph Convolutional Networks for user behavior prediction (jointly with DiDi AI Labs).
BSc, Computer Science (Artificial Intelligence) · King's College London2015 - 2018