RAG at Scale: Logging, Traceability, and the Architecture for Control

RAG pipelines are everywhere—but most are barely holding together. As GenAI moves from demos to production, the cracks are showing: silent failures, hallucinations, and a total lack of insight into what your AI is actually doing.

This talk introduces a new architectural mindset for building RAG systems that scale reliably and responsibly—with traceability and control built in from day one.

You’ll learn how to:

- Log retrievals, prompts, and responses in a way that exposes true decision lineage
- Trace outputs back to the documents—and even chunks—that influenced them
- Identify high-value patterns, failure clusters, and prompt blind spots over time

Use graph-native tools like Neo4j to map and monitor the system as it evolves

Whether you're running a few thousand queries or scaling to millions, this approach turns RAG into a system you can debug, explain, and trust. Because if AI is going to power real applications, we need more than answers—we need architecture.

Beyond Vectors: Evolving GenAI through Transformative Tools and Methods

Embark on a thought-provoking exploration of GenAI's evolution with "Beyond Vectors: Evolving GenAI through Transformative Tools and Methods." Tailored for engineers seeking fresh perspectives, this session encourages practitioners to step beyond familiar Vector Database practices. It's not just a departure; it's a pragmatic leap forward into precision methodologies for data quality and crafting datasets essential for Retrieval-Augmented Generation (RAG) excellence. We'll navigate the complexities of adding non-semantic context through graph databases, shedding light on the nuanced limitations of distance metrics like Cosine Similarity. Join us for this insightful journey, pushing the boundaries of GenAI evolution with transformative tools and methods.

Key Themes:

Methodical Precision in Data Quality and Dataset Construction for RAG Excellence: Uncover an integrated methodology for refining, curating, and constructing datasets that form the bedrock of transformative GenAI applications. Specifically, focus on the six key aspects crucial for Retrieval-Augmented Generation (RAG) excellence.

Navigating Non-Semantic Context with Awareness: Explore the infusion of non-semantic context through graph databases while understanding the nuanced limitations of the Cosine Similarity distance metric. Recognize its constraints in certain contexts and the importance of informed selection in the quest for enhanced data richness.

The Logging Imperative: Recognize the strategic significance of logging in the GenAI landscape. From application health to profound business insights, discover how meticulous logging practices unlock valuable information and contribute to strategic decision-making.

Key Takeaways:

Master a methodical approach to ensuring data quality and constructing datasets specifically tailored for Retrieval-Augmented Generation (RAG) excellence.
Navigate the complexities of adding non-semantic context, including an awareness of limitations in distance metrics like Cosine Similarity.
Understand the strategic significance of logging for application health and insightful business analytics.
Join us on this methodologically rich exploration, "Beyond Vectors," engineered to take your GenAI practices beyond the current Vector Database norms, unlocking a new frontier in GenAI evolution with transformative tools and methods!

Practical GraphRAG - Making LLMs smarter with Knowledge Graphs

We all know that LLMs hallucinate and RAG can help by providing current, relevant information to the model for generative tasks.

But can we do better than just vector retrievals? A knowledge graph can represent data (and reality) at high fidelity and can make this

rich context available based on the user's questions. But how to turn your text data into graphs data structures?

Here is where the language skills of LLM can help to extract entities and relationships from text, which you then can correlate with sources,

cluster into communities and navigate while answering the questions.

In this talk we will both dive into Microsoft Research's GraphRAG approach as well as run the indexing and search live with Neo4j and LangChain.

Identify Unknown Risk in Your Systems With Centrality Algorithms

How do you identify strengths and risk points in your complex systems? See how the Jedi and Rebel Alliance leverage graph data science for the forces of good.

Traditional risk management approaches often fail to capture the complexity of a system. The most impactful Empirical target may not be the most obvious target!

In this presentation, Alison and Jason will take the example of the Star Wars Rebel Alliance Network and demo with a Jupyter Notebook how to explore, clean up and analyze the Star Wars Galaxy with planets and hyperdrive lanes. She'll show you how you can use centrality algorithms to examine the most vulnerable Rebel planets, betweenness centrality to discover how to disrupt the supply chain, and pathfinding to navigate the optimal route.

Pattern Rights - An Ethical Framework for Generative AI Training Data

As generative AI continues to push boundaries, creating novel content by learning from massive datasets, we are faced with complex issues around intellectual property, privacy, and the ethical use of data. Current systems of copyright, fair use, and data protection lack the scope to fully address the unique challenges posed by AI pattern recognition and generation.

This pivotal talk introduces the pioneering concept of "Pattern Rights" - a holistic ethical framework to inform the development and deployment of generative AI technologies. Pattern Rights serves as an umbrella construct, encompassing principles of copyright, fair use, training data transparency, privacy, data ownership, and accountability.

We will explore how Pattern Rights can ensure appropriate attribution and compensation when AI models learn from copyrighted works or personal data. It establishes guidelines around consent, anonymization, and ethical data sourcing practices.

As the AI industry is rapidly evolving, we urgently need governance to foster innovation while upholding rights and safeguarding against misuse. Pattern Rights provides a roadmap to navigate this uncharted territory responsibly and equitably.

AI Evaluation: Tracing LLM Decisions for Reliability and Business Impact

As enterprises rapidly adopt LLMs for decision-making, they face a critical challenge: How do we evaluate and control AI-driven outcomes? Traditional AI monitoring tools only catch failures after they happen, but businesses need a way to trace, validate, and align LLM decisions before they cause financial or compliance risks.

This talk introduces graph-based AI evaluation—a method for mapping LLM decision pathways using Neo4j and Retrieval-Augmented Generation (RAG) to track data influence, improve model reliability, and ensure alignment with business goals. We will cover:

Why LLM decision failures happen—and why enterprises struggle to detect them early.
How graph-based AI evaluation helps businesses visualize AI decision logic, detect biases, and prevent costly mistakes.
Real-world applications of Graph AI in LLM deployments, including data-driven decision tracing and compliance monitoring.
LLMs are transforming business processes, but AI evaluation remains an unsolved challenge. This talk equips technical and business leaders with a practical framework for tracing AI decision-making, improving trust, and reducing risk.

The Hidden Patterns of Agentic AI—How Context Shapes Intelligence

AI agents don’t fail because of bad models—they fail because they follow the wrong patterns. Behind every hallucinated action or erratic tool call is an invisible structure: how agents retrieve, process, and decide in real time.

This talk pulls back the curtain on agent behavior patterns—the hidden loops, stale memories, and misaligned retrievals that quietly sabotage outcomes. You’ll learn:

* The 3 most common failure patterns in agentic AI—and why they’re hard to catch

* How agents lose context and fall into rigid loops (even in well-orchestrated chains)

* A blueprint for building adaptive agents using real-time behavioral context

You’ll leave with practical tools for making agents more reliable, interpretable, and self-correcting—without overhauling your stack.

AI’s Diversity Debt: How Compounding Bias Threatens Innovation—and What We Can Do About It

The tech industry is quietly amassing “Diversity Debt”—a hidden liability that arises from neglecting inclusivity, diverse data practices, and ethical governance in AI development. Like technical debt, Diversity Debt grows over time, with small biases in datasets, synthetic data generation, and model design compounding into systemic flaws that are increasingly costly and complex to address. These unchecked biases erode trust, stifle innovation, and create products that fail to serve the diverse populations they’re meant to empower.

In this keynote, Alison Cossette unpacks the concept of Diversity Debt and its far-reaching consequences, from skewed AI outcomes to diminished market opportunities. Drawing on her pioneering work in data provenance and governance with Neo4j and the FORGE platform, Alison demonstrates how organizations can identify and address compounding bias across all stages of AI development. Using compelling examples—such as synthetic datasets that reinforce disparities and feedback loops that amplify exclusion—she highlights the urgency of tackling bias before it scales.

Attendees will gain actionable insights into reducing Diversity Debt through inclusive governance frameworks, ethical data practices, and proactive model evaluation. Whether you’re a startup founder, data scientist, or industry leader, this talk will equip you to build AI systems that reflect the diversity of the world and unlock innovation without limits. Join us to discover why paying down Diversity Debt isn’t just ethical—it’s essential for creating AI that thrives in an increasingly complex, interconnected

The Black Box is a Lie: Why You Should Stop Blaming the Algorithm

It’s easy to call AI a black box, but that’s just an excuse for bad design. This session flips the script on opaque AI by exposing how human decisions—bad assumptions, shortcuts, and ignored data provenance—are the real culprits. Learn why transparency isn’t just an add-on but the foundation of ethical, accountable AI. This talk challenges participants to take back control from the “black box” myth and design systems that are clear, traceable, and human-centric.

None

Dynamic Data Intelligence: Enabling Proactive Governance and Risk Management in AI Systems

As artificial intelligence reshapes industries, the reliability, traceability, and risk associated with data take on unprecedented importance. Dynamic Data Intelligence (DDI) is an emerging competency of data governance that empowers organizations to understand, assess, and manage the origin, quality, and cascading impacts of their data across interconnected systems.

This session will explore how DDI enhances transparency and trust in AI workflows by dynamically identifying and prioritizing high-impact data. Through advanced modeling and analytical techniques, organizations can anticipate potential vulnerabilities, mitigate cascading risks, and optimize decision-making processes.

Key takeaways will include:

* Practical applications of DDI in pre-ingestion data governance to ensure the integrity of AI training data.
* Case studies showcasing how data risks were identified and managed across complex workflows.
* Metrics and methodologies for assessing data quality and understanding amplified risks.
* Real-world challenges in implementing dynamic data governance and how to overcome them effectively.

Join us to explore how Dynamic Data Intelligence transforms AI from a theoretical powerhouse into a practical, responsible, and sustainable solution, enabling organizations to achieve impactful AI innovations with confidence.

Building Physical AI: Fine-Tuning VLA Models and Composing Multi-Modal Systems

Physical AI requires a different stack than chatbots. You're fine-tuning vision-language-action models for specific skills, then composing them with perception and voice systems that work in real-time. The integration is where the magic happens—and where most projects fail.

This talk shares the architecture behind an award-winning book-reading robot: a system that opens books, turns pages, sees content, and reads aloud with expressive voices. We fine-tuned VLA models using Action Chunking with Transformers to learn fluid manipulation from human demonstrations, then integrated Claude Vision for page understanding and Eleven Labs for streaming speech synthesis.

I'll walk through the technical stack: how we collected demonstration data and fine-tuned for page-turning skills, why ACT policies outperform discrete skill primitives for manipulation, and how we achieved zero-latency speech through a three-threaded streaming pipeline. Physical AI means designing for graceful failure—because in the real world, errors tear pages.

You'll leave with concrete patterns for fine-tuning VLA models, composing multi-modal physical AI systems, and understanding where the hard problems actually live.

Learning Objectives
* Fine-tune vision-language-action models for specific manipulation skills
*Apply Action Chunking with Transformers for fluid robotic behavior
*Design multi-modal physical AI systems that integrate manipulation, vision, and voice
*Build zero-latency streaming pipelines for real-time human-robot interaction
* Identify failure modes unique to physical AI and design recovery strategies
Level
Intermediate

Tags
Physical AI, VLA Models, Robotics, Fine-Tuning, Vision-Language-Action, Real-time Systems

Will bring the robotic arm and can make available for demonstrations if interested.