Artificial intelligence is changing the way organizations operate, but one of the biggest questions leaders face remains the same: Should we build it ourselves or buy it? The answer isn’t just about cost. It impacts how quickly your organization can move, how adaptable your solutions are, who owns your data, and ultimately how you create a competitive advantage.
In innovative vehicle manufacturing, the fusion of Vision Systems and Artificial Intelligence (AI) has revolutionised quality control. As manufacturers pursue precision and efficiency, advanced vision technologies and AI algorithms ensure high production standards. Vision systems equipped with high-resolution cameras and lights serve as the vigilant eyes that meticulously scrutinise every aspect of vehicle assembly.
Governments are starting to use AI agents to improve how services are delivered—from permits and public works to citizen support. These “Civic Agents” are helping staff work more efficiently and making public services easier to access. In this session, we’ll look at real examples from local governments already using AI, and share a simple 5-step framework (Purpose, Pilot, Playbook, Production, Performance) for scaling these solutions.
As vehicle autonomy, connectivity and artificial intelligence reduce the level of control and increase exposure to attacks, it has never been more important to strengthen cybersecurity programs and be fully aware of emerging threats. Cybersecurity threats are not static; it’s a dynamic environment and constantly changing. Every day, there is a new potential threat, and while technology is becoming more and more sophisticated, actors are also becoming more and more intelligent and capable.
Modern enterprise AI is facing a silent production crisis: the "Instruction Drop." While Large Language Models (LLMs) appear highly capable in isolated demos, their reliability in complex, rule-heavy environments decays sharply once a prompt exceeds roughly eight core instructions. This session explores the technical "why" behind this failure—grounded in the "Lost in the Middle" phenomenon where transformer attention weights diminish at the center of dense prompts.
Large language models are only as useful as the context you give them. For data engineers, that context is your schema, your lineage graph, your query history, your dbt models — and most teams haven't connected those dots yet. This session is a practical introduction to Retrieval-Augmented Generation from a data engineering lens.
Your AI agent deleted 1,943,200 rows. Not because it was malicious — because in a documented March 2026 incident, a developer gave an agent a routine AWS cleanup task, it substituted terraform destroy for the instructed CLI commands without asking, and a production database vanished. No confirmation prompt fired. No override path existed. The model's guardrails didn't catch it.
The path from “I have a model” to “it’s serving production traffic on my cluster” is full of sharp edges. This talk is an end-to-end walk-through: GPU device plugins and node scheduling, model weight distribution strategies for 10–140 GB artifacts, health checks that survive multi-minute model loads, and GPU-aware auto-scaling. Includes a live demo deploying an open-weight LLM on Kubernetes and serving inference traffic with real-time GPU metrics.
As AI systems become increasingly integrated into enterprise applications, they also introduce a new and rapidly evolving attack surface. In this session, Ethan DeMott explores the emerging landscape of prompt-based attacks on large language models, including prompt extraction, prompt injection, and jailbreak techniques. Attendees will gain a practical understanding of how attackers exploit AI systems to leak sensitive data, manipulate outputs, and bypass safety controls.
Ever wonder how NASA writes software that literally can’t afford to fail? Meanwhile, back on Earth, our code is held together with duct tape and //TODO comments. The stakes might be a little lower for most of us—nobody's oxygen supply depends on our latest API deployment (hopefully)—but reliability, maintainability, and simplicity still matter. Enter NASA’s Power of Ten rules: a set of software engineering principles designed to minimize complexity and maximize reliability.
AI agents are rapidly becoming part of modern software, but many teams rush into building them without clearly defining the problem they’re solving or the role the agent should play. This session takes a practical, product management–driven approach to building AI agents that deliver real value. Instead of focusing only on tools or frameworks, it introduces a structured way to evaluate when agents are the right solution, and when they are not.
Introductions between buyers and suppliers for Oakland County Tech Collaboration members with a business presence in Oakland County, in partnership with OCTC and MEDC. Meetings are scheduled Thursday, October 29 and Friday, October 30, from 10 a.m. to 2 p.m. each day.
Artificial intelligence is changing the way organizations operate, but one of the biggest questions leaders face remains the same: Should we build it ourselves or buy it? The answer isn’t just about cost.
AI-powered vision can improve quality control only when it adapts to real production conditions. Quality systems have to respond to variation on the line, not just perform in a clean lab setup. "AI Influence on Reliable Machine Vision" lets Michael Koper connect vision data, AI, and manufacturing quality controls to the conditions where defect detection has to work.
Public agencies need AI that improves services without adding complexity. Governments are starting to use AI agents to improve how services are delivered—from permits and public works to citizen support. Himanshu Goil brings that problem into "Civic Agents: The Rise Of AI In Modern Public Service" by focusing on practical Civic Agent patterns for pilots, playbooks, and scale, not just the tool names around it.
Autonomous vehicles expand the attack surface across AI, connectivity, and control systems. Cybersecurity threats are not static; it’s a dynamic environment and constantly changing. In "Driving into the Future: Navigating Emerging Cybersecurity Threats in Autonomous Technology", Hemanth Tadepalli moves from the symptom to where emerging risks appear and how to reduce them, helping attendees see where better decisions start.
Enterprise prompts break down when too many rules compete for attention. The issue is not that LLMs are weak; reliability changes once instructions, policies, exceptions, and governance collide inside one workflow. In "Engineering Beyond the “Rule of Eight”", Harsh Ranjan turns that failure mode into architecture patterns for governed AI pipelines that hold up beyond a fragile prompt.
RAG is only as good as the metadata behind it. Large language models are only as useful as the context you give them. "RAG for Data Engineers: What It Is, Where It Fits, and Why Your Metadata Is the Missing Piece" keeps the conversation practical as Varun Joshi works through where schemas, lineage, query history, and dbt models fit in data engineering workflows.
AI agents can damage production data before a human sees the mistake. No confirmation prompt fired, no override path existed, and the database changed before anyone could intervene. In "Your AI Agent Just Ran DELETE Without WHERE. Now What?", Jonathan Stewart turns that failure mode into infrastructure-level gates that block destructive commands at execution time.
LLM features can become new attack surfaces. As AI systems move into enterprise applications, prompt injection, extraction, and jailbreaks turn from demos into real design risks. Ethan De Mott uses "Fortifying AI: Understanding Prompt Attacks and How to Defend Against Them" to connect those attacks to layered defenses teams can design before the feature is exposed.
Serving an LLM on Kubernetes has sharp edges. Includes a live demo deploying an open-weight LLM on Kubernetes and serving inference traffic with real-time GPU metrics. That is the entry point for "A Practical Guide to Running LLMs on Kubernetes", where Rohit Mishra shows how to handle GPU scheduling, model weights, health checks, autoscaling, and live inference in practice.
Synthetic USB models miss the messiness of real devices. Verifying USB Host designs-under-test (DUTs) is increasingly difficult as the USB ecosystem continues to expand across device types, protocol modes, and class behaviors. "Accelerating USB Host Verification Using Real USB Devices in a Virtualized Device Modeling Framework" turns that scenario into a session where Suchir Gupta shows how physical hardware can accelerate host verification in virtualized environments.
Digital tools stall when standard work and operators are disconnected. The idea only matters if it survives real equipment, operators, data quality, and production pressure. Bryan Bauw uses "From Standard Work to Scalable Smart Manufacturing: Making Data and AI Work in Assembly Operations" to show how MES, real-time data, and AI can scale assembly improvements without losing the human workflow they are meant to support.
Analytics teams need scale without losing flexibility or control. The session will also cover different ways to ingest data into Big Query and best practices, and will conclude with a live demo of the Big Query console and a showcase of a petabyte demo. In "Google Big Query - Heart of Smart Analytics toolset on Google Cloud", Vijaykumar Jangamashetti makes the problem concrete by showing how Big Query supports ingestion, ML, and petabyte-scale analysis on Google Cloud.
LLMs are powerful tools, but entity extraction still has classic CS problems. A tool can produce results while the underlying computer science problems are still very much alive. Milena F and Andy Schmidt use "LLMs are tools, not solutions: Entity extraction as a computer science problem" to show how context, ambiguity, and pipelines shape reliable results, giving attendees a practical way to bring the lesson back to their own systems.
The rapid scale‑up of lithium‑ion battery production for electric vehicles (EVs) and stationary energy storage systems (ESS) has introduced unprecedented challenges in leak detection and quality assurance. This paper presents a robotic tracer gas sniffing approach for high‑precision leak localization in EV battery packs and energy storage assemblies. The system combines industrial collaborative robotics with high‑sensitivity helium and hydrogen tracer gas detection to overcome the limitations of
Most production code is not going to space, but reliability still matters. Ever wonder how NASA writes software that literally can’t afford to fail? The answer gets practical in "Mission-Critical Code: What NASA’s Power of Ten Can Teach Us", where Jonathan "J." Tower shows how NASA's Power of Ten can simplify everyday software design.
Sepsis care depends on fast signals across fragmented workflows. The hard part is keeping trust intact as volume, ownership, definitions, and expectations change. In "AI and Digital Transformation in Sepsis Management for Smarter Healthcare Delivery", Swapna Chimanchodkar makes the next step concrete: how AI and predictive analytics can support earlier detection and response.
Production planning breaks when demand, supply, and capacity shift at once. Modern manufacturing environments operate under increasing complexity driven by demand variability, supply uncertainty, and capacity constraints. In "AI Driven Production Planning Using Decision Intelligence in Manufacturing Systems", Madhav Jayesh Kumar Pandya connects decision intelligence to the real-time tradeoffs behind better production planning.
Medallion architecture is harder under HIPAA. Bronze, silver, and gold layers are easy to sketch, but regulated healthcare data adds PII handling, lineage, access control, and audit pressure to every step. In "Medallion Architecture in Regulated Healthcare: A Hands-On Playbook", Raziullah Khan turns that architecture into patterns teams can use for cleaner, safer, audit-ready data work.
Modern products rely on operations customers never see. Delivery, reliability, data trust, and cloud spend all depend on disciplines that become visible only when one gap slows the business. In "The Invisible Stack: Black-Ops of Modern Enterprise", Robert Gray III maps Dev Ops, AIOps, Data Ops, and Fin Ops as one operating system for modern enterprise work, showing why the invisible stack matters before delivery, incident, or cost problems surface.
AI agents fail when teams start with the tool instead of the workflow. Once the idea reaches a real workflow, the hard questions shift to context, trust, cost, and reliability. Reshmika Dhandapani uses "AI Agents in Products: A Product Manager’s Guide to Building What Matters" to show how product managers define boundaries, value, and reliability.
5G promises low latency, but Android stacks can still bottleneck real-time apps. Real-time mobile experiences need the network, operating system, and edge architecture to work together. "5G-Optimized Android System Architecture for Low-Latency Applications" gives Patel Riddhi a way to make the issue concrete through architecture patterns for faster mobile and edge workloads.
Pipeline failures should not wait for someone to notice a bad report. Every data engineer knows the 2 AM pipeline failure — the one nobody notices until Friday's report is wrong. "Agentic Loops in the Data Stack: From Pipeline Failure to Auto-Remediation" turns that scenario into a session where Varun Joshi walks through agentic patterns for monitoring, root cause analysis, and auto-remediation.
Enterprise AI gets simpler when data and models work closer together. Oracle AI Database 26ai introduces a unified, enterprise‑grade foundation for building AI‑powered applications directly where your data lives. In "Building Enterprise-Grade AI Solutions with Oracle Generative AI Service and the Oracle AI Database", Juarez Junior moves from the symptom to the decisions underneath, showing how Oracle AI Database and OCI Generative AI support secure, scalable apps.
AI security changes when systems answer in probabilities instead of certainties. When I say "probabilistic" outcomes, your mind likely jumps to coins or dice. "Probably Secure: What We Misunderstand About AI And Determinism Is Making Us Less Secure" keeps the conversation practical as Dwayne Mc Daniel works through where deterministic controls still matter and where they fall short.
Data collection from industrial devices is often the least understood and most daunting task that a plant can face, yet having that data can give clarity on OEE, maintenance issues, utility usage, and many hard-to-spot trends. Temperature controllers, compressors, variable frequency drives, industrial cooling systems, and sensors that transmit data in any number of protocols (often archaic) will be explained and demystified in this session.
EV battery leaks are hard to find without slowing production. The idea only matters if it survives real equipment, operators, data quality, and production pressure. "Robotic Tracer Gas Sniffing for Leak Localization in EV Battery Packs and Energy Storage Systems" gives Davy Leboucher room to explain how robotic tracer gas sniffing improves localization across complex battery assemblies.
Plant data is valuable, but industrial devices and old protocols make collection intimidating. The first useful step is often less about dashboards and more about getting trustworthy signals out of the equipment already on the floor. Grant Pinkos uses "How to get started with collection and storage of industrial data" to make that starting point concrete, from collection and storage to analysis and alerts.
Artificial intelligence is changing the way organizations operate, but one of the biggest questions leaders face remains the same: Should we build it ourselves or buy it? The answer isn’t just about cost. It impacts how quickly your organization can move, how adaptable your solutions are, who owns your data, and ultimately how you create a competitive advantage.
