Alejandro Araujo Rajzner
Azure AI Architect | AI governance for Microsoft-based enterprises | Microsoft Certified Trainer
Zürich, Switzerland
Actions
Alejandro Araujo Rajzner is an independent Azure AI Architect and Microsoft Certified Trainer based in Zürich. He helps organisations move AI agents from pilot to production with governance built in from day one: EU AI Act and ISO/IEC 42001 readiness, data governance with Microsoft Fabric and Purview, and secure agent architectures with Copilot Studio, Microsoft Foundry and MCP.
Before going independent, he spent 11 years at Microsoft as a Senior Cloud Solution Architect, working with more than 200 enterprise customers across financial services, pharma and energy. Today he trains and coaches Microsoft employees and global partners at L200 to L400 and authors hands-on labs for Skillable.
He speaks regularly at the Microsoft AI Tour, MCT Summit, Experts Live and community events across Europe and the Middle East. He serves on the AI CERTs Advisory Board and writes Guardrails First, a vendor-neutral AI governance newsletter.
Area of Expertise
Topics
ISO 27001 + ISO/IEC 42001 en Microsoft 365: gobernar Copilot y agentes con evidencias, no con promes
La IA generativa ya está dentro del tenant, y las auditorías ISO 27001 empiezan a preguntar por ella. ISO/IEC 42001 añade el sistema de gestión de IA encima del SGSI. En esta sesión se conectan ambas normas con lo que Microsoft 365 ya ofrece: plantillas de Compliance Manager, Purview DSPM for AI, etiquetas de sensibilidad y DLP para Copilot, Entra para identidades de agentes y Defender para detección. Demo práctica: mapear controles del Anexo A de 27001 y de 42001 a configuraciones reales, y montar la carpeta de evidencias que un auditor pedirá sobre Copilot y un agente de Copilot Studio. Se cierra con los errores más comunes al extender un SGSI existente a la IA y con una hoja de ruta de 90 días.
Te llevas: matriz de controles 27001/42001 frente a herramientas M365 y un checklist de evidencias.
Guardrails First: EU AI Act Obligations for GCC Organisations Deploying Copilot and Azure AI
Any GCC organisation with European customers, subsidiaries or partners is already inside the scope of the EU AI Act, and most have not noticed. Meanwhile agents are multiplying across Copilot Studio and the wider Microsoft stack faster than governance can follow. This session is the practical version: what the AI Act actually obliges you to do, how those obligations map onto a Microsoft agent estate, and where Microsoft's agent control plane fits. We take one agent from build to classification, work through the risk tier decision live, and produce the documentation an auditor would ask for. You leave with the classification method and the evidence template, not a compliance lecture.
Takeaways:
- Which EU AI Act obligations reach a GCC organisation, and which do not
- A risk classification method applied end to end to one Microsoft agent
- An evidence template you can use on your own agent estate
Del token al crédito: FinOps real para escalar proyectos Microsoft GenAI
La adopción de Microsoft GenAI (Azure OpenAI, Copilot, agentes, RAG, orquestaciones) está creciendo más rápido que la madurez en la gestión de costes. Muchos proyectos empiezan con pilotos controlados… y escalan sin una metodología clara, hasta que el director financiero pregunta por qué el coste ya no se mide en licencias, sino en consumo impredecible.
En esta sesión presentaré un enfoque práctico y accionable de FinOps aplicado a proyectos GenAI en Microsoft, explicando cómo crear tu propia metodología para gobernar costes desde el diseño, no cuando llega la factura. Analizaremos cómo cambia el modelo mental al pasar de coste por usuario/licencia a coste por token, throughput, latencia y créditos, y qué implicaciones tiene esto para arquitectura, uso y operación.
Se profundiza en la transición del coste del token al modelo de créditos, cómo interpretarlo correctamente, y cómo conectar este consumo con decisiones técnicas: elección de modelos, tamaño de prompts, RAG vs fine‑tuning, caching, reintentos, agentes encadenados y patrones de uso. Veremos cómo introducir guardrails financieros, presupuestos, alertas y chargeback/showback sin frenar la innovación.
La sesión propone una metodología FinOps específica para GenAI, alineada con prácticas Microsoft, que cubre:
Diseño (cost‑aware architecture),
Implementación (instrumentación y medición),
Operación (control, optimización y forecasting),
Gobernanza (roles, responsabilidades y conversación con Finanzas).
El objetivo es claro: permitir que equipos técnicos y de negocio escalen GenAI de forma sostenible, con visibilidad, control y un lenguaje común con el área financiera.
Qué se llevarán los asistentes:
Un framework FinOps específico para Microsoft GenAI (no genérico cloud).
Cómo pasar de “coste desconocido por IA” a coste explicable y gobernado.
Patrones y anti‑patrones reales que disparan el consumo de tokens/créditos.
Claves para alinear arquitectura, uso y Finanzas sin bloquear la adopción.
¿Cuántos agentes hay en tu tenant? Inventario y gobierno de agentes
Copilot Studio, agentes de SharePoint, agentes declarativos, conectores MCP: los agentes se crean más rápido de lo que se gobiernan. Sin inventario no hay gobierno, ni clasificación de riesgo para el AI Act, ni respuesta cuando alguien pregunta qué datos toca cada agente. Esta sesión muestra cómo descubrir y catalogar los agentes del tenant, asignar propietario y nivel de riesgo, aplicar políticas de entorno y DLP en Power Platform, y controlar su ciclo de vida. Basada en implantaciones reales con equipos de partners y clientes enterprise, con demo de principio a fin.
Te llevas: plantilla de inventario de agentes y criterios de clasificación de riesgo.
Your Copilot Studio Agent Looks Great. Have You Actually Evaluated It?
Everyone's building Copilot Studio agents. Almost no one evaluates them before sharing. In this fast, demo-first session we take a customer-support agent that sounds confident and on-brand... then run Copilot Studio's built-in Agent Evaluation against it. Watch the groundedness score expose that our "great" agent is quietly inventing return windows, lifetime warranties, and discounts that don't exist. Then we harden it and watch the scores jump, live. You'll leave knowing exactly how to evaluate your own agents this week, it's a CSV and a button.
RAG but for tools: Meet Foundry Toolbox
Retrieval-Augmented Generation transformed how AI systems access knowledge. But what if the real bottleneck isn't finding information—it's finding, securing, and governing the right tools?
As organizations build increasingly capable AI agents, tool sprawl becomes a major challenge. Teams create duplicate integrations, credentials are scattered across applications, governance becomes inconsistent, and every new agent requires yet another round of tool wiring. Microsoft Foundry Toolbox introduces a different approach: curate tools once, expose them through a single MCP-compatible endpoint, and let agents dynamically discover and use the capabilities they need.
In this session, we'll explore why Foundry Toolbox can be thought of as "RAG for tools", a centralized, governed, and discoverable layer for APIs, MCP servers, search services, code interpreters, and other agent capabilities. You'll learn how Toolbox enables reusable tool catalogs, centralized authentication, policy enforcement, versioning, and seamless consumption from any MCP-compatible agent framework.
Through practical demos and architecture walkthroughs, we'll show how to reduce integration complexity, improve governance, and accelerate agent development by connecting agents to a single endpoint instead of dozens of individual services. Whether you're building enterprise copilots, multi-agent systems, or AI-powered applications, you'll leave with a new mental model for tool orchestration and a blueprint for scaling agent ecosystems without creating tool chaos
AI - Ethics, Morals, Purpose & Identity: Impact of Job Displacement
When AI reshapes or removes a job, the loss is rarely just income. It's identity, confidence, routine, and purpose.
Over three months, a small international group convened by Lesley Crook (MVP/MCT) debated exactly this, producing more than 300 pages of insight on career reinvention, digital confidence, ethics vs morals, and the difference between a machine that simulates understanding and a human whose judgment is shaped by lived experience, empathy, and common sense.
This panel shares what the group learned and then goes one step further. You'll see the first live version of the "Career Renewal Adviser", an agent built in Copilot Studio that starts from a person's whole story rather than their CV and skills, and surfaces transferable strengths and lived experience, mapping realistic pathways after AI-driven displacement.
El agente decidió. Ahora explícalo tú.
Tu agente de Copilot está en producción. Responde correos, resume reuniones, busca en SharePoint, dispara flujos. Nadie lo para. Nadie lo supervisa. Y el día que haga algo que no debía, te van a llamar a ti.
En esta sesión vamos al fondo del problema: qué ocurre cuando un agente de Copilot Studio actúa de forma inesperada, por qué los permisos de SharePoint que nadie revisó son ahora un riesgo real, y qué herramientas tienes hoy mismo en el stack de Microsoft para no llegar a esa conversación con Legal sin respuestas.
Sin humo. Sin demos de catálogo. Casos reales, decisiones reales, y las preguntas que tu organización debería hacerse antes de que alguien se las haga por ti.
Lo que te llevas:
Un modelo mental para evaluar el riesgo de un agente antes de desplegarlo
Los controles concretos que tienes en M365 hoy: Purview, sensitivity labels, permisos de Graph, Copilot Studio
Las preguntas exactas para las conversaciones difíciles con stakeholders no técnicos
Guardrails First: A Practical Playbook for AI Risk, Compliance & Privacy in the Microsoft Ecosystem
Organizations are deploying Copilot and AI agents faster than they’re governing them, and with the EU AI Act’s August 2026 enforcement deadline approaching, the cost of “figure it out later” just got real. This session delivers a practical, Microsoft-native playbook for building layered guardrails across four dimensions: data classification and access boundaries, AI interaction controls, privacy risk management, and audit-ready oversight. You’ll see how Purview, Priva, Defender for Cloud, and Entra map to specific failure modes, and leave with a rollout checklist that proves control without killing velocity.
From Copilot to Agents: Turning Daily Work into Measurable Outcomes
Many organizations have started with Microsoft 365 Copilot but still treat it as “a smarter search box.” This session shows business and project leaders how to turn those early wins into end-to-end outcomes powered by agents. We’ll walk through three tangible patterns:
Customer email → triaged, summarized, and routed by an agent;
Sales opportunity → notes, next steps, and proposal draft prepared automatically;
Internal request → approval flow with an agent doing the heavy lifting.
For each, we’ll map the business owner, data needed, risks, and KPIs (time saved, cycle time, quality). You leave with a small playbook of patterns you can take straight into your next customer or internal project.
Experts Live Emirates 2026 Sessionize Event Upcoming
Microsoft 365 Live 2026 Sessionize Event
Copilotverse Global User Group User group Sessionize Event
CollabDays Madrid 2026 Sessionize Event
North America MCT Summit 2026 Sessionize Event
AICD - Dubai 2026 Sessionize Event
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top