Agenda

Building a RoboTaxi platform taught me that traditional AI architectures crumble under the complexity of large-scale multi-agent communication.

In this talk, I'll share hard-won lessons from architecting a distributed AI system where hundreds of autonomous agents coordinate thousands of rides daily. You'll discover why point-to-point agent communication creates a brittle mess, and how event-driven architecture (EDA) becomes the nervous system that keeps everything connected.

Through a live demo of the RoboTaxi platform, I'll show you:

- How AI agents can communicate asynchronously without tight coupling

- The architecture decisions that separate toy demos from production-ready AI systems

Whether you're building chatbot networks, autonomous systems, or multi-agent AI applications, this session will give you the architectural foundation to scale beyond the prototype phase.

Let's take this ride together.

At Microsoft, we run AB testing in pre-production to catch regressions before release. But figuring out whether an incident was a true positive, a false alarm, or a missed regression used to mean hours of manual detective work.

In this talk, I’ll share how we built ALFA (Automated Loop Feedback Analysis), a system that uses Large Language Models (LLMs) to close the feedback loop between incidents in production and gaps in pre-production.

You’ll discover:

• How LLMs extract structured signals (builds, error codes, event tables) from messy incident reports.
• Why reasoning-based binary classification beats rule-based heuristics for identifying missed regressions.
• How we use LLMs for root-cause analysis — spotting coverage gaps, signal gaps, and process gaps — and automatically generating recommendations.

Many others rush into AI adoption without internal coordination or responsible guardrails, which could not only waste time and resources but also pose security risks and destroy clients' trust. The answer to both situations is an ethical and effective AI governance. In this session, let’s talk about practical approaches to govern AI across different industries and settings:

-        Setting priorities, choosing use cases for AI, and building an AI strategy

-        Establishing AI governance and audit, improving privacy, security and compliance with the EU AI Act

-        Learning from success and scaling from silos to impact

more predictable, and more resilient leveraging on agentic and classical ML. 

In this session, I will discuss the essential factors and considerations for successfully moving AI systems from prototype to a production-ready state.

The pace of innovation has shifted from evolution to acceleration, moving from systems that follow instructions to systems that understand context, remember intent, and collaborate through tools.

We’ve built astonishing tools, but not always for the right reasons.

This talk challenges the innovation race and asks the question few dare to: “Should we?” instead of “Can we?”.

From ethical design to long-term impact, we’ll explore how technology becomes truly meaningful when built with purpose, empathy, and intention.

Expect stories that inspire, provoke, and maybe unsettle, about what happens when humans remember why we build in the first place.

Because the next era of technology won’t be defined by power, but by purpose.

This hands-on session shows how to build agentic systems with local-first SLMs, either via Ollama server using OllamaSharp or fully in-process with ONNX. You’ll see how SLMs deliver lower latency, reduced action costs, stronger data control, and ease of deployment for edge or on-prem environments. Architecture is exemplified with live demos of real-world processes, highlighting tool use, state management, and safe recovery loops. Scenarios include autonomous execution and human-in-the-loop approvals/escalations. Attendees will leave with concrete deployment patterns, working C# examples, and practical guidance for making SLM-first agentic AI cost-effective and production-ready.