XY

Senior Research Fellow · AIDF, NUS Singapore

XuanYao姚璇

I build AI systems for finance, learning, and research.

My work sits at the intersection of financial AI, trustworthy evaluation, and adaptive learning—with a focus on language models, multi-agent systems, and credit risk. I also lead the AI Tutor program.

Explore AI Tutor Google Scholar
Portrait of Xuan Yao
Senior Research Fellow / AIDF, NUSProgram Lead / AI Tutor
Financial LLMsCredit riskMulti-agent systemsAI tutorsInterpretable AI
AI Tutorflagship research program for grounded, adaptive learning
EMNLP2026 Main Conference paper accepted
40+junior researchers, students, and interns mentored
2international Best Paper Awards

01 / Flagship project

AI Tutor, from research to practice.

AI Tutor is the core research program. FinTutor is its flagship financial-education application, translating that research into a system for real learners and institutions.

01

AI Tutor

Research program

A reusable learning infrastructure that connects grounded AI tutoring, adaptive practice, learning analytics, and content governance.

Grounded tutoring · Adaptive practice · Learning analytics
02

FinTutor

Flagship application

The financial-education application of AI Tutor, co-developed with insurers and professional examination bodies for trusted learning at scale.

Adaptive learning · Financial education · RAG guardrails
↗

02 / Research themes

The questions behind
the systems.

These are long-running research directions rather than standalone projects or individual publications.

A

Financial AI evaluation

Benchmarks for reasoning, factuality, citations, calculations, temporal validity, latency, and cost.

B

Credit risk intelligence

Models and benchmarks spanning disclosures, events, sentiment, and corporate networks.

C

Deep research agents

Evidence-grounded agents for financial research, live forecasting, and decision support.

D

Adaptive learning systems

Grounded tutors that adapt practice while remaining legible, governed, and safe.

03 / Selected research

A thread of ideas.

Each item below is a specific publication, accepted conference paper, or methodological contribution.

2019

NeurIPS · Time-series forecasting

Enhancing the Locality and Breaking the Memory Bottleneck of Transformer on Time Series Forecasting

An early exploration of Transformer architectures for long, fine-grained time series, introducing convolutional self-attention and a memory-efficient LogSparse pattern.

Read the paper
2026

HICSS

Corporate Default Prediction Through Text Mining: Integrating Event, Sentiment, and Network Analyses

Corporate default · Financial NLP

2026

AAAI Workshop · Oral

Fin-Agent-QG: A Collaborative Multi-Agent Framework for Higher-Order Cognitive and Difficulty-Aware Financial Exam Question Generation

Multi-agent systems · Financial education

2025

ICAIF Workshop · Oral

Evaluating Large Language Models for Financial Reasoning: A CFA-Based Benchmark Study

LLM evaluation · Financial reasoning

04 / Questions I’m thinking about

Notes from the edge of the work.

  1. 01

    How do we evaluate financial AI when the ground truth is incomplete, delayed, or market-dependent?

  2. 02

    Can an AI tutor adapt deeply without becoming overconfident or opaque?

  3. 03

    What should a research agent do when the right answer is “there isn’t enough evidence yet”?

About this lab

I’m a Senior Research Fellow at the Asian Institute of Digital Finance, National University of Singapore, and Program Lead for AI Tutor. My work connects financial reasoning, credit-risk modelling, interpretable AI, and learning systems.

This site is a living index of the projects, papers, tools, and questions that make up that work. For research and collaboration, reach me at yaoxuan@nus.edu.sg.