Daniel Deng

Vietnam's banking leaders have stopped asking abstract AI questions. They're focused on the operational reality. After our roundtable, three challenges kept surfacing for scaling AI safely and driving real ROI: 1. Execution: How do we move from chatbots to Agentic AI that executes? 2. Governance: How do we deploy AI with speed, yet maintain regulatory discipline? 3. Trust: How do we embed AI without damaging customer and employee trust? AI only delivers when data, architecture, and human oversight work together. Access the full strategy here: https://lnkd.in/gWGxHBz6

90

1 Comment

I'm delighted to announce the launch of the Capital One and University of Illinois Urbana-Champaign (UIUC) Center for Generative AI Safety, Knowledge Systems and Cybersecurity (ASKS). This collaboration deepens our partnership with UIUC and expands our partnerships with top tier universities to drive cutting-edge AI research and solve substantive, industry-wide challenges. In line with Capital One's approach to mission-inspired science, the aim of ASKS is to advance the state of the art in AI while addressing the challenges and opportunities of Generative AI models in today's quickly evolving world. In addition to research, education and the responsible advancement of Gen AI models across sectors and domains, the Center will provide support to selected faculty research projects and PhD students through fellowships focused on Gen AI model research and safety, and will host a research-led symposium later this fall. Read more: https://lnkd.in/gP76xd6T University of Illinois Urbana-Champaign Heng Ji Milind Naphade Sambit Sahu Bayan Bruss Stephen Rawls Elizabeth Boschee Daben Liu Erin Madden Jennifer Borden Gang Wang Gokhan Tur Nancy Amato

474

7 Comments

🚀 Generative AI is transforming industries—but we also have to govern it with the rigor it demands! In my latest article for DATAVERSITY, I lay out a practical framework for responsible adoption of generative AI. Whether you're shaping strategy in the boardroom or building models through engineering teams, governance is no longer optional it's foundational. 🔍 Key mechanisms I explore in the article: ▶️ Clear decision rights & accountability ▶️ Dynamic policy frameworks & monitoring ▶️ Risk assessments & AI registries ▶️ Multi-line defense operations Alignment with NIST & ISO 42001, geographical and industry standards 💡 This piece is for anyone navigating the fast-moving AI landscape from executives and architects to analysts and regulators. 📖 Read the full article, I’d love to hear how your organization is approaching generative AI governance. What’s working and what’s still a work in progress? https://lnkd.in/djPh-XVu #GenerativeAI #AIGovernance #ResponsibleAI #ISO42001 #NISTAI #DataLeadership #TechEthics #DATAVERSITY #AIstrategy #DigitalTransformation

35

9 Comments

One of the most transformative innovations powering today’s GenAI revolution is the advancement of foundation models (FMs) — large models pre-trained on Internet-scale data that demonstrate remarkable generalization across a wide range of tasks. FMs have redefined performance benchmarks in many fields including, natural language processing, computer vision, and computational biology, becoming the cornerstone of modern GenAI. Yet, progress on structured enterprise data such as tabular time series has lagged behind, where task-specific models optimized on manually extracted signals still dominate. Inspired by the success of FMs in other domains, we set out to explore whether FMs can also be built for finance and banking. After over three years of R&D, I’m excited to share that we’ve deployed our first structured-data FM into production at TD: TD AI Prism https://lnkd.in/gEAb7Scm. Trained on billions of customer events, Prism generalizes to many downstream predictive tasks, delivering higher accuracy and significantly faster time to market. Through this work, we're beginning to observe early signs of scaling laws, similar to those seen in LLMs, where more data and larger models lead to stronger generalization. As part of this effort, we also released a public FM trained on Internet tabular data, capable of making predictions on new datasets without any fine-tuning. Check it out here: https://lnkd.in/gmbWdAf7 As enterprises evolve with AI, we believe that foundation models for structured data will be a major force in redefining how predictive AI is deployed at scale, driving the next wave of innovation in enterprise applications. Layer 6 AI has multiple exciting initiatives in this space, so stay tuned for more updates!

360

15 Comments

I’m often asked: "How do you effectively scale AI across an organization?" Scaling AI means responsibly maturing from proof-of-concepts to production-ready capabilities that become integral to an organization’s operational fabric. For best results, it requires democratizing AI access across all organizational levels and functions. At RBC, AI is empowering our people, helping uncover opportunities to better serve and understand our clients and enabling productivity gains that help our teams focus on higher-impact work. some of the areas we’re focusing on: 1. Platform and Foundation Success requires robust capabilities that are compliant, cost-effective, and convenient (aka user-friendly). Organizations must establish well-defined, reusable patterns for AI integration. Without standardized platforms and frameworks (at RBC this is our Lumina platform), teams can sometimes reinvent processes, creating challenges in scalability, security, and governance. 2. Access and Education AI accessibility must encompass diverse employee roles. RBC prioritizes AI fluency and trains our leaders and employees on how to work with and integrate AI responsibly into their roles. We also encourage a "learn by doing" mentality and will support employees on this AI journey in the years ahead. 3. Culture and Adoption Perhaps most critically, scaling AI requires fostering a culture of openness, curiosity, and continuous learning. Success demands organizational commitment and sustained motivation to integrate AI into daily workflows. At RBC, our development teams use AI for code generation and system enhancement, over 29,000 employees have been onboarded to RBC Assist, our in-house developed genAI tool, to help improve productivity and efficiency. It's incumbent on all of us to learn how to work with AI and leverage these tools to amplify our impact. Our AI adoption is aligned with our organizational goals to foster a culture of curiosity where change readiness is an everyday ability. It's evident that AI will fundamentally reshape how we work and interact with technology and the ability to scale AI effectively will be a critical driver of organizational success in the future. #scaleAI #borealis #lumina #rbcassist

258

3 Comments

AI Engineering: A Year of Disruption and Opportunity: Thank you to Phillip Radley for hosting our February 2026 episode of Data & AI Book Club on architecting and building AI Powered Applications. We explored the rapid evolution of AI engineering, using Chip Huyen O’Reilly book as a launchpad. A few highlights and check out the book for a whole lot more on the biggest change in our lifetime! 🧰 🔹 AI Engineering Is Evolving Fast: The discipline is rapidly expanding beyond traditional ML, with new frameworks, agentic architectures, and solution patterns emerging. Get hands on with tools. Staying current is essential. 🔹 Foundation Models & Context Are Critical: Understanding how foundation models work, their biases (especially toward English), and techniques like RAG and fine-tuning is vital for building effective AI solutions. 🔹 Vibe Coding Is Changing the Game: Coding is becoming outcome-driven; consumer and enterprise AI platforms let you build apps by simply describing what you want, opening up new possibilities for rapid prototyping and innovation. 🔹 Open Weight Models & Global Competition: The rise of open weight models (especially from China) is democratizing access to powerful AI, challenging US dominance and reshaping the global AI landscape. 🔹 Paradigm Shift - Build vs. Buy: AI is disrupting the economics of software - organizations can now build custom solutions quickly and cheaply, challenging the traditional SaaS model and prompting us to reimagine what is possible in how we approach enterprise tech. Our Westpac Data & AI Book Club is our member-driven community for curious, continuous learners and Data & AI enthusiasts. What new skills do you think are most important for AI engineers in the agentic era? 👇 #DataBookClub #Westpac #AIEngineering #AI #ArtificialIntelligence #Architecture #Engineering

35

Enterprise AI Systems Architecture Blueprint 🔰 A Strategic, Modular, and Scalable Framework for Enterprise AI Agent Systems This blueprint presents a governed, composable, and observability-ready architecture for deploying AI agents in enterprise environments. ⸻ 🧱 I. Core Functional Layers (Vertical Execution Stack) Each layer has a distinct, non-overlapping responsibility, aligned with modular engineering and governance principles: Layer Purpose 1. UX & Interface Layer Human-AI interaction: text, voice, visual UI. Connected via API Gateway. 2. Integration & Interoperability Layer API and service integration with internal/external systems. 3. Trust, Ethics & Safety Layer. Responsible AI: HITL, bias detection, safety nets. 4. Meta-Orchestration Layer Agent coordination, multi-agent planning, planner-critic loops. (Insert above Principle Layer) 5. Principle Layer Zero Trust, least privilege, modular boundaries, enforcement logic. 6. Cognitive Intelligence Layer Reasoning, reflection, long-term memory, system intelligence. (Consolidate duplicates) 7. LLM Engineering Layer Prompt chaining, embeddings, retrieval, token management. 8. Agent Execution Layer Tool/function calling, decision logic, feedback loop. 9. Platform Layer (MCP) Model routing, fallback, caching, token monitoring, MCP Gateway. 10. Infrastructure Layer. Compute (Kubernetes), GPUs, networking, Istio, service mesh, observability stack. ⸻ 🔄 II. Cross-Cutting Concerns (Applied Across All Layers) Applied horizontally to every layer to ensure governed, optimized, and secure AI operations. Domain Details Governance Policies, ethical frameworks, audit, responsible AI Security Encryption, identity, RBAC, secret management Optimization Caching, cost tuning, latency control Observability Distributed tracing, logs, metrics, SLOs ⸻ 🛠️ III. Systemic Support & Enablement Layers (Foundational Stack) Layer Description 1. Routing Layer Model orchestration, versioning, gateway routing, A/B testing 2. Deployment Layer Containerisation (Docker, K8s), CI/CD pipelines, rollbacks 3. Non-Functional Layer SLA/SLO, disaster recovery, HA, fault tolerance ⸻ 🧠 IV. Architectural Principles Principle Meaning 🎯 Separation of Concerns Each layer isolates a responsibility for resilience and clarity. 🧩 Composable Modularity Replaceable agents, APIs, and logic blocks. 🔐 Secure by Design Built-in least privilege, zero trust, and encryption. 🧠 AI-Native Enablement Supports autonomy, reasoning, memory, and orchestration. 🏢 Enterprise-Grade Integrates with observability, governance, and cloud-native practices.

99

3 Comments

Learn how a strategic partnership between applied research and product management allows Capital One to drive scientific innovation at scale and create a better banking experience for our customers. Read the full blog from Bayan Bruss, VP, Applied AI Research, and Ofer Idan, Sr. Director, Enterprise AI/ML Product, on our strategy-backed applied research. https://bit.ly/4o8oh62

5

I have been digging into how we evaluate large language models, especially when they’re used in decision-making workflows like fraud detection, onboarding, or compliance. In Part 1 of my series, I look at the current benchmarks (MMLU, TruthfulQA, HELM) and why they often miss what matters most- hallucinations, reasoning quality, and context consistency. High scores don’t always mean high trust. Especially when the model confidently makes up a policy clause. If you’re working with LLMs beyond chat use cases, I’d love your take. https://lnkd.in/g8cuqQAj #LLM #AIevaluation #generativeAI #ResponsibleAI #Compliance #GenAI #fintech #banking

25

1 Comment

“AI agent evaluation replaces data labeling as the critical path to production deployment” perfectly captures something many of us are starting to feel in real projects. Now that companies are building complex, reasoning-capable agents, the real challenge isn’t the data anymore — it’s how we evaluate the agents themselves. How do they behave in the wild? Are their actions aligned with our goals, policies, and risk frameworks? That’s the new hard challenge. I’d love to hear from others who are deep into this: How are you approaching agent evaluation? Are you using frameworks, synthetic testing, or human-in-the-loop validation? Let’s share what’s working — this might be one of the key challenges shaping the next generation of production AI. https://lnkd.in/dbsMMAJn

30

3 Comments

Major microstructure changes and rough dates that you should know when backtesting with CME market data: Jan 2, 2013 - Messaging Efficiency Program becomes effective. Jun 3, 2013 - Self-match prevention functionality launched. Mar 9, 2014 - New iLink and MSGW architecture (Phase 1) launched for TAS markets. Sep 28, 2014 - CME introduces MDP 3.0 based on SBE. Sep 21, 2015 - BTIC launched. Jan 31, 2016 - New iLink and MSGW architecture pre-launch for remaining markets (Phase 2). Apr 10, 2016 - New iLink and MSGW architecture in production for remaining markets (Phase 2). Jul 10, 2016 - Major performance enhancements to CME Globex. (?) Jul 17, 2016 - Production rollout of iLink nanosecond timestamp enhancement and tag 5979. May 21, 2017 - Production rollout of MBO for final market segment. Jul 17, 2017 - Russell index futures and options return to CME. Sep 24, 2017 - Implied functionality for FX futures and SD calendar spreads. Dec 18, 2017 - CME launches crypto futures with Bitcoin futures. Mar 25, 2018 - FX Link launched. May 13, 2018 - TACO launched. Jul 29, 2018 - iLink performance enhancements. Jan 13, 2019 - 9-digit price precision extension. Feb 3, 2019 - Launch of consolidated iLink fill messages. May 6, 2019 - Micro E-mini futures launched. Jun 17, 2020 - Production rollout of binary iLink 3.0 for final market segment. Jun 27, 2021 - Major performance enhancements to CME Globex. Jun 27, 2021 - 3:15 - 3:30 PM CT pause eliminated for CME equity index futures and options. Sep 17, 2021 - Standard-sized S&P 500 index futures and options delisted. Apr 14, 2023 - Eurodollar futures and options converted to SOFR contracts. Jun 26, 2023 - Eurodollar futures and options delisted with expiration of June contracts. Jan 7, 2024 - Major performance enhancements to CME Globex. Jan 7, 2024 - Mass quote messaging capacity increased. Mar 25, 2024 - Micro Treasury futures launched. Need a microstructure-aware vendor and market depth going back to 2009? You get both with #Databento.

164

8 Comments

This is interesting—Google is releasing a new T5 model. Before the term "GenAI" was popularized and Decoder-only models became dominant, the Encoder-Decoder architecture was the standard. T5 models, in particular, enjoyed widespread popularity a few years ago. https://lnkd.in/gYgy8MRG

8