About

Current

I am an ML Research Scientist at Sapient Intelligence in Beijing. I am developing Praxit, an autonomous AI-scientist framework with DAG-based execution lineage for reproducible solution provenance. On Karpathy’s AutoResearch prompt, Praxit achieved roughly 30% higher performance at 10% of the cost of Claude Code with Opus. I also generate synthetic time-series data and develop HRM-based forecasting models for weight-management applications.

In 2026, I showcased Praxit at the AI4 conference through live demonstrations and conversations with prospective customers and collaborators.

Research

Based at the University of Cambridge and collaborating with Shanghai Jiao Tong University, I developed PLASMA, an optimal-transport module for fast and interpretable protein substructure alignment. It improved ROC-AUC by 10–30% across seven protein-language-model backbones and three VenusX tasks, and was accepted as a poster at ICLR 2026.

At Cambridge, I developed TCPNet, an SE(3)-equivariant topological neural network that jointly learns residue- and secondary-structure-level protein features without losing geometric information (invited talk). The project received the Department of Computer Science and Technology’s Highly Commended M.Phil Project Prize 2024–2025. I also worked on multi-omics integration with graph neural networks and GraphAU-Pain, presented at the IJCAI 2025 MiGA Workshop.

At UCL, I collaborated with clinicians from UCL Hospital on readmission prediction using remote patient monitoring data. I developed a SQL-based data pipeline, performed clinical feature engineering, and built interpretable models with a ROC-AUC of 0.79 ± 0.03 and SHAP analysis.

Education

At the University of Cambridge, I completed an M.Phil. in Advanced Computer Science with Distinction, ranked 8th out of 60. Courses included Geometric Deep Learning, Affective AI, Mobile & Wearable ML, NLP, and Multi-agent RL.

I graduated with First-Class Honors in B.Sc. Computer Science from University College London. Courses included Machine Learning, Computer Vision, Reinforcement Learning, Mathematics & Statistics, and Software Engineering.

Industry Experience

At Xunfei Healthcare Technology, I built a knowledge-distillation pipeline using a teacher LLM to synthesize medical instruction examples, improving the Xiaoyi student model’s instruction-following accuracy from 34% to 82% through supervised fine-tuning.

At Luojin Data Information, I developed a RAG pipeline with a MongoDB vector store and LangChain agents for automated financial-report classification, achieving 92% query-translation accuracy and a 97% entity-matching rate.

I also collaborated with IBM and the NHS to create an automated chatbot-generation service for hospitals, reducing development time from weeks to 30 minutes. The project was demonstrated live at Great Ormond Street Hospital and presented to IBM, Microsoft, and Intel.

Academic Service

I am a reviewer for NeurIPS 2026 and for the AI4Science Workshops at NeurIPS 2025 and ICML 2026. I was also a Programming Tutor at UCL in 2022–23.

Research Interests

I am interested in AI agents, large language models, geometric deep learning, and AI for biology and healthcare. I welcome opportunities to collaborate on technically ambitious, high-impact work.

Technical Skills

Machine Learning: PyTorch, PyTorch Lightning, PyTorch Geometric, NumPy, Pandas, Scikit-learn

Languages: Python, JavaScript, Java, C++, R

Research Infrastructure: SLURM, Weights & Biases, TensorBoard