ACM SIGKDD 2026 · Lecture-Style Tutorial

Generative AI & Large Language Models
for Financial Markets

From Behavioral Prediction to Autonomous Trading

A three-hour journey through the parts of generative AI that actually survive contact with markets — behavioral foundations, grounded LLM systems, and responsible deployment.

Zangir Iklassov Hachem Madmoun Jean-Jacques Duhot Salem Lahlou

August 9–13, 2026

Jeju Island, Republic of Korea

3 hours · half day

Explore the schedule Slides & notebook

Hours, three acts

Live demonstration

Hands-on notebook

Tutors across research & industry

A different view of finance & AI

Financial markets are one of the few machine-learning settings where feedback is fast, mistakes are expensive, and deployment is constrained by execution and regulation. Yet most existing tutorials still treat finance either as an NLP benchmark or as a classical forecasting problem.

This tutorial takes a different view. We show how recent progress in generative AI becomes genuinely useful in markets when it helps model participant behavior, ground reasoning in live documents and market data, and support research and execution workflows that can survive contact with production.

Behavioral & Market Foundations

Framing markets as behavioral systems, and the anatomy of a systematic trading strategy — from labeling and feature research to meta-models and portfolio construction.

Grounded LLM Systems

Retrieval-augmented generation, critique models, and agentic workflows — with hallucination control framed as evidence alignment, not prompt engineering.

III

Responsible Deployment

Leakage-aware backtesting, execution and risk realities, and the governance questions that arise once these systems leave the lab.

Tutorial schedule

A research-to-deployment progression. Timings include a live demonstration, a real-strategy case study, and short Q&A — total 180 minutes.

What is being modeled, and which data sources carry behavioral information.

15′

Framing markets as behavioral systems

Why modeling how participants process and react to information is often more decision-relevant than treating price as an isolated target.
Iklassov
35′

The anatomy of a systematic trading strategy

From raw prices to a live futures strategy: labeling, feature creation, feature importance, side-and-size meta-models, CPCV validation, and portfolio construction.
Madmoun Case study · the Hendaye strategy at Alken Asset Management
10′

Bridge to financial LLMs

From early disclosure-based text analysis to timely, grounded, uncertainty-aware extraction from earnings calls, filings, tables, and charts.
Lahlou

From individual signals to systems that ground, compare, and act on them.

25′

Retrieval-augmented generation for market intelligence

Long-context architectures, retrieval, self-reflective RAG, factuality, calibration, and uncertainty — auditable at inference time.
Lahlou Live demo · end-to-end RAG on an earnings call
20′

Agentic AI for autonomous trading

How research automation, monitoring, and decision support change once models can plan, use tools, keep memory, critique, and update hypotheses.
Iklassov
15′

Multi-agent collaboration & conflict resolution

Disagreement between agents treated as measurable evidence about uncertainty, model diversity, or regime change — with human-in-the-loop escalation as a design feature.
Iklassov & Lahlou

Which ideas survive realistic validation, execution constraints, and governance.

25′

Rigorous backtesting

Look-ahead bias, survivorship bias, selection effects, and cross-sectional contamination — with purged cross-validation, embargo periods, and multiple-testing corrections.
Madmoun
10′

Risk management & execution realities

Why spread, market impact, funding liquidity, queue position, volatility targeting, and drawdown controls must be designed in from the start.
Madmoun
25′

Moderated ethics & governance discussion

Three disagreements that arise precisely when systems are useful: predation vs. discovery, systemic fragility, and transparency vs. competitive advantage.
All presenters

Tutors

Researchers working on reasoning, uncertainty, and agentic AI, together with practitioners and mentors rooted in systematic trading and optimization.

Presenters

Zangir Iklassov

Assistant Teaching Professor & Research Scientist, MBZUAI

In-person · Corresponding tutor

Zangir works on reinforcement learning for combinatorial optimization, LLM reasoning, and agentic AI for sequential decision making — including a NeurIPS 2024 paper on self-guiding exploration for combinatorial problems. Before joining the faculty he led research for ADNOC's EnergyAI initiative and co-founded Smart System Technologies, and he has given invited talks including at the University of Oxford. He leads the behavior-first framing, agent design, and multi-agent coordination modules.

Salem Lahlou

Assistant Professor of Machine Learning, MBZUAI

In-person

A core contributor to Generative Flow Networks (GFlowNets), Salem's research spans uncertainty quantification, sample-efficient reinforcement learning, generative models, and language-model reasoning. He completed his PhD at Mila / Université de Montréal in Yoshua Bengio's group, and previously worked at the Technology Innovation Institute, Booking.com, Google, and IBM. He leads the financial NLP, RAG, prompt design, and confidence-aware evaluation sections.

Hachem Madmoun

Co-founder, Syllogia · Lecturer, Imperial College Business School · AI Researcher, MBZUAI

In-person

Hachem works at the interface of AI research and systematic asset management. He is the co-founder of Syllogia, a research company that serves as an external pod in the systematic team of Alken Asset Management. He also lectures on systematic trading and machine learning in finance at Imperial College Business School, and is an AI Researcher at MBZUAI. He holds a PhD in applied mathematics from École des Ponts ParisTech, where he developed regime-detection models for asset allocation and stock-picking strategies, and co-authored the FinChain benchmark for verifiable financial reasoning.

Special Collaborators

Mentors and practitioners who shaped the tutorial's design, optimization grounding, ethics, and governance content.

Martin Takáč

Associate Dean for Academic Affairs & Associate Professor of Machine Learning, MBZUAI

Special collaborator

A leading researcher in large-scale optimization for machine learning — spanning stochastic, distributed, and federated optimization — Martin's work (including the widely cited SARAH gradient method and reinforcement-learning approaches to combinatorial routing) has drawn over 11,000 citations, and he regularly serves as an Area Chair for NeurIPS, ICML, and AISTATS. Previously a professor at Lehigh University, he holds a PhD in mathematics from the University of Edinburgh. His guidance and support were instrumental in shaping this tutorial.

Jean-Jacques Duhot

Senior Portfolio Manager, Brevan Howard Asset Management LLP

Special collaborator

Jean-Jacques brings decades of institutional experience across discretionary and systematic investing in commodities and cross-asset macro. He founded and served as CIO of Arctic Blue Capital — a systematic manager later majority-acquired by H2O Asset Management, where he became deputy CIO — following senior roles at Société Générale, CDPQ, and Millennium. His collaboration focuses on the tutorial's execution and risk-management material.

Who it's for & what you'll learn

Designed for

Machine-learning researchers seeking an application domain with measurable consequences
Quantitative researchers updating their toolkits for the generative-AI era
Practitioners and policy researchers needing a realistic view of these systems in markets

Prerequisites

Supervised learning, neural networks, optimization, and working Python (PyTorch, scikit-learn, pandas), plus basic familiarity with liquid markets and buy/sell orders. No prior LLM or quantitative-finance background is required.

By the end, participants can

Reason about behavior-first signal design
Build and critique RAG and multi-agent pipelines for market intelligence
Evaluate them with leakage-aware backtesting
Discuss the execution, governance, and ethical questions that arise once these systems leave the lab

Societal impact, treated as design

Market stability, fairness, compliance, and accountability are framed as design constraints — model diversity, stress testing, kill-switch logic, and human oversight — situated within the EU AI Act, MiFID II, ESMA guidance, and evolving SEC expectations.

Materials

The slide deck and hands-on notebook for the session.

Slide deck

The full lecture deck for all three parts, combined in one PDF.

PDF · 233 slides

Notebook · Earnings-call RAG

End-to-end RAG on an earnings call, from retrieval to a confidence-scored signal.

Jupyter notebook

Cite this tutorial

@inproceedings{iklassov2026genaifinance,
  author    = {Iklassov, Zangir and Madmoun, Hachem and
               Duhot, Jean-Jacques and Lahlou, Salem},
  title     = {Generative AI and Large Language Models for Financial
               Markets: From Behavioral Prediction to Autonomous Trading},
  booktitle = {Proceedings of the 32nd ACM SIGKDD Conference on Knowledge
               Discovery and Data Mining V.2 (KDD '26)},
  year      = {2026},
  publisher = {Association for Computing Machinery},
  address   = {Jeju Island, Republic of Korea},
  doi       = {10.1145/3770855.3816459}
}

Generative AI & Large Language Models
for Financial Markets