The One About OCR (ft. OCBC AI Lab & Standard Chartered)
What does reliable document extraction look like beyond traditional OCR?
Hear how practitioners are building multistage pipelines for complex financial documents and evaluating the growing use of VLMs for OCR. From retrieval and structured extraction to benchmarking, get a closer look at how modern document AI systems are built and tested for real-world use.
More About the Sharings
Yuxuan & Yuanxing (Data Scientists, OCBC AI Lab) will share on “A Multistage Extraction Pipeline for Long Scanned Financial Documents”
Long financial documents are messy: scanned pages, multiple languages, inconsistent layouts, and only a small amount of information that matters. While vision-language models can process these documents end-to-end, doing so reliably in real-world KYC workflows remains challenging.
Yuxuan and Yuanxing will share how they developed a multistage extraction pipeline combining image preprocessing, multilingual OCR, page-level retrieval, and VLM-based structured extraction. Drawing on experiments across 120 production KYC documents and around 3,000 scanned pages, they'll explore why finding the right pages before asking a VLM to reason over them can significantly improve extraction accuracy, and what their ablation studies reveal about building document AI systems for real-world financial workflows. (Technical Level: 200)
Dr. Yiming Qian (Senior Scientist, A*STAR) and Dr Huang Jie (AI Research Scientist, Standard Chartered Bank) will share on “OCR With VLMs: Evaluation Framework”
Vision-language models are increasingly being used for OCR and document understanding, but headline accuracy alone doesn't always reflect how well these systems perform in real-world workflows. Some fields, such as payment amounts, account numbers, dates, or compliance information, carry far greater consequences when extracted incorrectly.
Drawing on applied work with Standard Chartered Bank, Dr. Qian will introduce a business-oriented framework for evaluating VLM-based OCR. Explore how traditional accuracy metrics can be complemented by critical-field-weighted evaluation, robustness testing, and failure analysis to better understand where models perform well, where they break down, and which errors matter most in practice. (Technical Level: 100)
More About the Speakers
Yuxuan Han is a Data Scientist at OCBC Group Data Office, where she works on applying machine learning and advance analytics to support business operations and compliance initiatives. Her research interest includes large language model and efficient model architecture, particularly in model compression and pruning strategy. Her work has been published at ICLR 2026, and ACL 2026 (Industry Track). Her work at OCBC focus on building scalable data-driven solutions and deploying machine learning models in real-world financial use cases.
Yuanxing Zhang is a Data Scientist at OCBC Bank’s AI Lab, where she builds production NLP and agentic AI systems for financial crime, payments, and corporate banking, including KYC document extraction, name screening, and LLM-powered operational workflows. Her work has been published at ACL 2026 (Industry Track) and a KDD 2026 workshop, representing OCBC in document intelligence and explainable money mule detection. She holds a B.Comp. in Computer Science (AI) from NUS and is currently pursuing an M.Sc. in Data Science and Machine Learning.
Dr. Yiming Qian is a Senior Scientist at Singapore’s A*STAR, specializing in AI, generative AI, and enterprise AI systems. He also works closely with Standard Chartered Bank on applied GenAI initiatives. With experience spanning research, financial services, and entrepreneurship, Dr. Qian has led the development of large language model applications, multi-agent AI systems, intelligent workflow automation, and AI-driven decision-support solutions. He previously co-founded an AI technology company that was acquired in 2020 and continues to advise and build AI ventures. Dr. Qian holds a Ph.D. in Computer Science from York University, Canada, and his current interests focus on transforming advances in AI into scalable, reliable, and commercially valuable systems for enterprises.
Huang Jie is AI Research Scientist at Standard Chartered Bank. She holds a PhD from Nanyang Technological University (NTU), where her research focused on molecular dynamics simulations. Her career spans software engineering, AI consulting at Boston Consulting Group (BCG), and applied AI in financial services. At Standard Chartered, she has worked on AI solutions for financial crime detection and now focuses on AI Research & Experimentation, exploring how emerging AI technologies can be translated into practical banking applications. She also supports research collaborations and partnerships across Asia with universities, research institutes, and industry partners.
More About The Series
AI Wednesdays is Lorong AI’s weekly gathering, bringing together practitioners, researchers and innovators for technical discussions on research insights, product development and engineering practices.
Get involved: Learn more about Lorong AI | Speaker Sign-up | WhatsApp Community | LinkedIn | X