----------------------------------------------------------------------------------------------------

 

 

 

           Department of Systems Engineering and Engineering Management

 

 

 

                                   The Chinese University of Hong Kong

 

 

 

----------------------------------------------------------------------------------------------------

 

 

 

Date: Friday, July 10, 2026, 16:30pm to 17:30pm HKT

Venue: ERB513, The Chinese University of Hong Kong

Title: From LLMs to AI Agents: Building Financial Intelligence That Can Reason, Plan, and Act

Speaker: Dr. Austin Zhang, Capital One

 

Abstract: 

Large Language Models are rapidly changing how financial institutions serve customers, automate workflows, and build intelligent products. In this talk, we will share how we are developing next-generation AI systems for complex financial reasoning and real-world customer assistance.

We will introduce MACAW - Multi-Agent Conversational AI Workflow, an LLM-based framework designed to go beyond simple chat. MACAW enables conversational assistants to understand user intent, reason through complex financial questions, plan multi-step actions, generate precise API calls, and reflect on their own outputs for better reliability. Within Financial Services, MACAW has been integrated into customer-facing applications such as Customer Assist, where it helps deliver API-grounded, business-logic-aware, and action-oriented customer experiences.

The talk will also provide a behind-the-scenes view of how we build Financial LLMs end to end, including pretraining, supervised fine-tuning, direct preference optimization, and reinforcement-learning-style training such as GRPO. We will discuss the technical challenges of building reliable AI agents in high-stakes financial domains, and highlight open research and engineering opportunities for students interested in LLMs, agents, reasoning, post-training, and applied AI.

 

Biography:

Dr. Zhang is the VP of LLM core team at Capital One, where he leads the LLM pretraining, post-training and architecture research. Prior to his current role, he was a Principal Lead of the multi-modal LLM and Speech team at Tencent AI Lab. Austin Zhang obtained his Ph.D. from the University of Cambridge in 2014. Following his PhD studies, he served as a Sr. Speech Scientist at Microsoft, where he built Microsoft Speech Recognition Models serving billion's customers in 200+ languages. Dr. Zhang's scholarly contributions were honored with the Best Paper Awards at Interspeech and NAACL, and he received the "IC Greatness Award" at Microsoft for his crucial role in developing the "Personalized Hey Cortana" system for Windows 10. He frequently gives keynotes/tutorials and has served as Area Chair (ICASSP, Interspeech, ASRU) and elected vice-chair of IEEE Speech Language Technical Committee (SLTC).