Video by FINOS via YouTube
Richard Harmon (Global Head of Financial Services at Red Hat) breaks down the fast-approaching frontier of financial engineering. From Barclays’ agentic trading desks to diffusion foundation models trained on exchange tick data, Richard explores how central banks and tier-one institutions are transitioning past narrow transformer correlations toward causal "World Models" and resilient AI architectures.
🗽 Catch Us in New York! Ready to explore the future of agentic AI, world models, and financial resiliency? Join the global banking community at OSFF New York on November 4–5, 2026.
🎟️ Register Now: https://hubs.ly/Q04n_bZL0
🔥 20% OFF DISCOUNT CODE: 26YTOSFFNY20C
🕒 Timestamps:
0:00 Introduction: Global Financial Trends & Open Source Investment
1:11 $5 Billion Security Commitment & The Emerging Tech Spectrum
3:30 Case Study: Barclays’ Agentic Trading & Analyst Swarms
4:18 Calculating Risk via LLMs: Estimating Stochastic Differential Equations
5:05 Foundation Models for Exchange Data: Moving Beyond Text
6:06 Auto-Regressive vs. Diffusion Models for Limit Order Books
7:46 The Limits of Transformer Architectures & Compounding Forecast Errors
8:20 World Models: Moving from Correlation to Causal AI
10:00 Operational Resiliency: The Bank of England Definition
11:21 Transitioning from Experimentation to Industrialized Agentic Systems
12:32 Shifting Left on AI Governance, Ethics, and Model Validation
13:50 Agility & Architecture: Why Platforms Must Evolve Every Six Months
📊 The Problem: The Forecasting Error of Auto-Regressive Models
Standard Large Language Models (LLMs) rely on auto-regressive processing—evaluating sequence data token-by-token or tick-by-tick. In high-frequency capital markets or complex risk calculations, a single minor prediction error compounds sequentially down the execution chain. This creates massive forecast errors, making pure transformer-based architectures unreliable for modeling real-time market risk, limit order books, or complex portfolio exposures.
🏗️ The Solution: Causal World Models & Diffusion Architectures
Richard Harmon maps out the next generation of financial AI infrastructure:
Diffusion Foundation Models: Evaluating billions of potential market outcomes holistically rather than sequentially, allowing the system to self-correct prediction errors instantly.
Causal World Models: Shifting AI from basic statistical correlation to true environmental context and causality, enabling agents to operate with structural awareness.
Agentic Risk Guardrails: Structuring "agent-checking-agent" review loops (as demonstrated by Barclays) to validate stochastic calculations against historical baselines before market execution.
⚙️ Why This Matters for Financial Engineering
Safe Sandbox Testing: Capital markets serve as an optimal testing ground for agentic AI because workloads do not expose private retail customer data while offering immense quantitative leverage.
Regulator-Aligned Resiliency: Designing systems to meet central bank expectations (like the Bank of England’s operational resilience framework) by building automated recovery paths directly into the architecture.
🌐 More about FINOS: https://www.finos.org/
📧 Join our newsletter: https://www.finos.org/sign-up
🎙️ Listen to our Open Source in Finance Podcast: https://www.youtube.com/@FINOS/podcasts
LinkedIn: https://www.linkedin.com/company/finosfoundation
#FINOS #OSFFNewYork #RedHat #CapitalMarkets #AgenticAI #WorldModels #QuantFinance #DiffusionModels #RiskManagement #FinTech