Data Scientist at OCBC AI Lab, working on Document AI and NLP. Student of Nanyang Technological University, Major in Computer Control & Automation, My research interests include Medical Imaging, Computer Vision, Natural Language Processing and Parallel Computing.

Click here to see where I’ve been for School and Lab visiting.

Education

• B.S. in Information Engineering
  Jilin University, China, 2016 - 2020
  Supervisor: Jian Wei
• M.Sc. in Computer Control and Automation
  Nanyang Technological University, Singapore, 2021 - 2022 Supervisor: P. N. Suganthan
  Co-supervisor: ArulMurugan Ambikapathi
• Exchange
  Department of Electrical and Computer Engineering
  University of Miami, Jul.2018
  Supervisor: Prof. Xiaodong Cai
• Lab Visiting
  • Built Environment City Analytics Laboratory
    UNSW,
  • Manufacturing Technology Research Laboratory
    University of Manchester,
  • Autonomous System and Robotics Lab
    Salford Innovation Research Centre(SIRC)
  • Robotics Innovatory
    Sungkyunkwan University
  • Department of Electronic and Communication
    Far Eastern Federal University

Intern

• Algorithm Engineer Intern
  PanoAI, Chengdu, Sep 2020 - Dec 2020

• Algorithm Engineer Intern
  Machine Intellection I2R A*Star, Singapore, Jun 2021 - Feb 2022

Work

• Data Scientist
  OCBC AI Lab, Singapore, Jun 2022 - present
  • Productionized VLM-based document extraction pipelines (invoice, term sheet, classification) on OCBC LLM Cluster with an end-user UI, rolled out company-wide; ~92–95% accuracy across 3 business units.
  • Sole contributor in designing and deploying a vLLM/SGLang + LiteLLM inference stack on GPU cluster, enabling day-1 LLM/VLM model support for internal business units company-wide.
  • Led OCR-based key field extraction from 2M+ historical KYC scanned documents using Qwen3.6 and PaddleOCR, contributing structured data to support a Client Lifetime Management system.
  • Built NLP model services for Group Legal & Compliance (NER-based term labelling, 5+ FTEs saved) and Treasury Operations (MiniLM SWIFT message classification integrated with PEGA, 3 FTEs saved).
  • Built a Python package for Singapore and Malaysia reconciliation process, onboarding 300+ term sheet templates across 5 product types, processing 1k+ trades monthly.

Publications

• A Multistage Extraction Pipeline for Long Scanned Financial Documents: An Empirical Study in Industrial KYC Workflows
  ACL 2026 (Industry Track), pp. 1419–1433
• Perception, Layout, and Validation: Calibrated Confidence for Reliable Straight-Through Processing of Financial Documents
  DocInsights Workshop @ EMNLP 2026, pp. 232–243
• Image Fidelity and VLM Ensemble for Expert-Level Single-Page Document Parsing on Dr.DocBench
  DocInsights Workshop @ EMNLP 2026, Shared Task Paper
• Multi-Stage Field Extraction of Financial Documents with OCR and Compact Vision-Language Models
  arXiv preprint, 2025

Competitions

• 3rd Place, Dr.DocBench Challenge, EMNLP 2026 DocInsights Workshop, Oct 2026
  Overall score 85.63, with an image-fidelity and VLM-ensemble approach to expert-level single-page document parsing.
• Top 50, GIS Image Registration Competition, 2020
  Participated during the PanoAI internship.
• ICCV 2019 (9th place) and Alibaba Tianchi NER Challenge (Top 0.5%): see the ICCV and Tianchi posts.

Research Projects

• Image/Text Style Transfer and Disentanglement
  College of William and Mary, Apr 2021
  VAE-based approach using mutual information and PCA to calculate and standardize hidden-variable encodings for image/sentence feature disentanglement.

Recent News

Attended The 2nd Large-scale Video Object Segmentation Challenge, ICCV2019.