Video by FINOS via YouTube
Dr. Jochen Papenbrock (EMEA Head of Financial Technology at NVIDIA) breaks down Jensen Huang’s 5-Layer AI Factory stack and explores how open-source foundation models power production-grade financial applications. From Royal Bank of Canada’s multi-modal equity research agents to Revolut’s custom transaction foundation models, Jochen details the open microservices, agent toolkits, and guardrailing frameworks needed to scale trustworthy AI in regulated banking.
🗽 Catch Us in New York! Ready to build scalable, trustworthy AI infrastructure? Join us 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: Deconstructing Jensen’s 5-Layer AI Factory Stack
1:03 The AI-Powered Bank: Post-Training Customization & Intelligence Flywheels
2:32 RBC Case Study: Accelerated Multi-Modal Deep Research Workflows
3:49 Next Frontier: Training Transactional Foundation Models from Scratch
4:46 Revolut & Payments Industry Blueprint: Tabular Transformers in Action
5:35 The Flywheel Engine: NeMo Microservices & NeMo Guardrails
7:02 The Nemotron Model Family: High-Efficiency Reasoning and Agent Harnesses
8:24 Managing Agent Fleets: NeMo Agent Toolkit & Profiling Latency Gains
9:28 Deep Research Blueprints: Equity Analysis Architecture
10:54 NeMo OpenShell: Trust Boundaries and Policy Signing Gateways
12:20 Summary: TCO Optimization, Token Manufacturing, and GPU Acceleration
📊 The Problem: The High Friction of Siloed Financial AI Systems
Deploying point-solution AI models without a cohesive platform leads to massive hardware inefficiencies, high latency, and data privacy vulnerabilities. Financial institutions attempting to run agentic workflows across fragmented infrastructure struggle with high token costs, difficult regulatory compliance tracking, and an inability to continuously capture operational learnings back into their systems.
🏗️ The Solution: The 5-Layer AI Factory & NeMo Ecosystem
Jochen Papenbrock provides a full-stack blueprint to operationalize AI inside banks:
Custom Post-Training & Weights Control: Utilizing open-source foundation models (like Nemotron) where banks retain access to model weights, enabling internal domain tuning for high-precision tasks.
NeMo Microservices & Guardrails: Deploying open-source runtime checks, benchmarking tools, and GPU-accelerated policy enforcement layers to secure agent interactions.
Tabular Transaction Transformers: Moving beyond text-based LLMs to train custom transactional foundation models from scratch using tabular customer behavior data.
⚙️ Why This Matters for Financial Engineering
Open Shell Security Boundaries: Enforcing model-agnostic trust boundaries and policy signing gateways around every agent action in production.
Optimized Token TCO: Co-designing software microservices directly with underlying GPU hardware to maximize token generation throughput while minimizing energy and infrastructure costs.
🌐 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 #NVIDIA #NeMo #Nemotron #AIFactory #CapitalMarkets #TabularTransformers #GPUAcceleration #FinTech #JochenPapenbrock