Secure and Responsible AI: A Zero Trust Approach to LLM-Powered SQL Systems

Today, organizations are increasingly using LLM to allow users to query data stored within the enterprise domain using natural language. There are a number of advantages to doing so, but a number of security risks are involved. These include prompt injection, attempts to jailbreak the application, unintended access to database schema, and SQL injection. These risks cannot be addressed by conventional means and thus emphasize the need to use a Zero Trust approach, wherein all data is validated prior to accepting it.

This session will demonstrate a proven approach to securing a natural language to SQL application.

Architecture Overview

The solution will use a multi-layered approach, wherein each layer will have a specific role to play. The layers are as follows:

UI Layer (React):
This is the frontend of the system. The system is designed to use natural language to accept the query entered by the user. The query is then forwarded to the backend services.

API Layer (.NET Core):
This is the second layer of the system. The API layer handles the security of the system. It is responsible for authenticating and authorizing users. Only users who are authorized can access the system through this layer.

LLM Function App (Python 3.10 – Azure Functions):
This is the main component of the system. The component interacts with the LLM. The LLM is responsible for generating SQL queries. The security validations of the system are performed in this layer. The system then runs the query on the database and presents the output to the users.

Security Design (Zero Trust in Practice):
Instead of having a single point of security check, the system has security checks at multiple points. The security checks of the system occur at the following points:

1. Input Validation (Pre-LLM):
Before sending the query to the LLM, the system validates the user input. The system utilizes the following tools to validate the user input:

* Azure AI Content Safety Prompt Shields
* Custom Python Validation Logic

2. Controlled SQL Generation
The system generates the SQL queries using LangChain’s SQldBToolkit, which is integrated with the ChatGPT model. However, the system does not execute the queries directly.

Instead, the system performs the following operations on the query:

* Query Rewriting
* Allow-list validation of tables, columns, and operations

3. Secure Query Execution
The system only allows the execution of validated queries on the SQL database. The system does not execute the query directly from the LLM.

4. Output Validation (Post-Execution)
After executing the query, the system scans the output set for sensitive data, including:

* Personally identifiable information
* Salary or financial data
* Internal schema data

If the system detects any sensitive data, it blocks the output.

5. Structured Output Parsing
The system utilizes a custom output parser, which parses the output into the following formats:

* JSON
* CSV
* HTML

This ensures the system does not expose the output of the database.

6. Logging and Monitoring
The system utilizes tools like Azure Monitor or Logspace to log all the actions, including security-related actions. The system logs the following data:

* Blocked Queries
* Detected Threats
* Policy Violations

Key Takeaway
If we connect the LLM model to the database directly without proper security measures, it is dangerous. To make it safe, we need to ensure that the model is validated at every stage of input, processing, execution, and output.
This architecture shows how to safely allow natural language access to enterprise data and how to control security and data exposure.

Breaking the Black Box: Production-Grade Distributed Tracing for MCP with OpenTelemetry

MCP servers are black boxes. Your agent calls a tool – what happens inside? How to debug slow calls, silent failures, or erratic behavior?

This session delivers a production-ready blueprint for distributed tracing in MCP with OpenTelemetry.

Part 1 – What, why, when: Understand traces, spans, and why MCP uniquely needs them.

Part 2 – Technical deep dive: MCP is transport-agnostic (stdio, HTTP, WebSockets). Learn how the _meta field carries W3C Trace Context across JSON-RPC – enabling end-to-end tracing. Implementation patterns in Python, TS, and Go.

Part 3 – Business value: Faster debugging reduces MTTR, audit trails support compliance (SOC2, HIPAA, GDPR). We reference OWASP MCP08:2025 on audit & telemetry.

Part 4 – Live demo: Instrumented MCP server + Jaeger. See developer debugging, performance bottleneck isolation, and business audit trails in action.

Learn MCP tracing mental model (_meta, transport-agnostic). Get ready code (Python). Master best practices (sampling, spans). Discover open tools (Jaeger, SigNoz,). Build business case (MTTR, compliance). Start today with checklist & demo.

Building Trustworthy ML Workflows: Hands-On with RAI Dashboards and Azure ML Pipelines

Building machine learning models is only the beginning. In this hands-on session, you’ll learn how to build trustworthy and explainable ML workflows using Azure Machine Learning Pipelines and Responsible AI (RAI) Dashboards.
Through a practical end-to-end example, we’ll automate data preprocessing, model training, and responsible model evaluation using Azure’s RAI tooling — all orchestrated through a scalable, modular pipeline.
Whether you're new to MLOps or aiming to embed responsible AI practices into your production workflows, this session will help you turn best practices into reality.

Key Takeaways:

Understand how to design trustworthy, explainable ML workflows using Azure ML Pipelines.

Learn to automate data preprocessing, model training, and responsible model evaluation with RAI tools.

See how scalable, modular pipelines can embed Responsible AI principles across the ML lifecycle.

Gain hands-on experience leveraging Azure ML Studio and Azure AI Foundry ecosystem services.

Discover practical strategies for integrating fairness, explainability, and error analysis into production-grade ML systems.

Implementing Responsible AI on Azure: Governance, Compliance, and Guardrails for Best Practices

As organizations race to adopt AI solutions, the need for ethical guardrails has never been more critical. With mounting regulatory scrutiny and growing public concern, implementing responsible AI isn't just good practice—it's becoming a business imperative.
Drawing from my experience implementing Azure AI solutions across multiple industries, I'll share practical approaches to embedding responsibility into every stage of the AI lifecycle. This isn't about theoretical frameworks that look good on paper but fall apart in production. It's about real, implementable strategies that maintain compliance without sacrificing innovation.
Attendees will walk away with:
✅ Governance Frameworks for AI – Concrete steps to establish responsible AI policies and oversight mechanisms that actually work in complex enterprise environments
✅ Compliance with Global Standards – Navigating the maze of GDPR, CCPA, and emerging regulations without stifling innovation or getting lost in legal speak
✅ Fairness & Bias Detection – Hands-on techniques using Azure's Fairness tools to identify and mitigate bias before deployment, with special focus on high-risk domains like healthcare and finance
✅ Explainability & Transparency – Making the "black box" transparent with Azure's explainability features, including how to communicate model decisions to stakeholders and end-users
✅ Privacy & Security – Practical approaches to data minimization, differential privacy, and federated learning that protect sensitive information without compromising model performance
✅ Best Practices & Real-World Case Studies – Lessons learned from implementing responsible AI at scale.

I'll demonstrate these concepts with live examples and provide attendees with a comprehensive toolkit they can implement immediately. My session will conclude with a roadmap for organizations to assess their current Responsible AI maturity and establish clear next steps.
As responsible AI transitions from aspiration to regulation, developers need practical guidance now more than ever. Join me to ensure your Azure AI implementations don't just meet technical requirements, but stand up to ethical and legal scrutiny.

Hands-on with Azure SQL binding for Azure Functions

Azure Functions can now interact with SQL databases through input and output bindings for Azure SQL or On-premises SQL. With Azure SQL bindings, data can be input from a database to the function with an input binding and data can be output from the function to the database. Configuring these bindings is accomplished with a small number of parameters, including the specification of the database query or table and the connection string.

In today's session, we will build a simple Azure Function for both SQL Input and Output bindings.

Build hybrid apps with .NET MAUI

In this session, we'll explore why a hybrid web approach is useful for building native apps with .NET MAUI in .NET 9. We'll have a hands-on session with Blazor hybrid, showing how to create native apps for Windows, Mac, iOS, and Android with a single codebase using web skills. We'll also look at the new HybridWebView control and how to use other web frameworks within .NET MAUI.

AI-Powered Frontend Development: From Copilots to Component Generation

In this dynamic session, we'll explore how AI is revolutionizing frontend development, from intelligent code completion to complete component generation. Learn how to leverage AI tools to accelerate your development workflow, improve code quality, and stay ahead in the rapidly evolving frontend landscape. Through live demonstrations and practical examples, you'll discover how to effectively integrate AI assistants into your daily development process.

Key Learning Outcomes:

* Master prompt engineering techniques for optimal results with GitHub Copilot and similar AI assistants
* Learn to efficiently generate and customize components using AI tools like v0.dev
* Understand best practices for reviewing and optimizing AI-generated code
* Develop strategies for successful team adoption of AI development tools

FDA-Compliant Review Workflows with Power Platform: Ensuring 21 CFR Part 11 Compliance

In this presentation, we'll explore the creation of FDA-compliant approval workflows within the Microsoft Power Platform, with a specific focus on meeting the stringent requirements of 21 CFR Part 11. Our journey will encompass:

1. Developing a streamlined approval workflow in Power Automate
2. Constructing a custom approval component using the Power Platform Component Framework
3. Deploying and demonstrating this component in action

21 CFR Part 11 is a crucial FDA regulation governing electronic records and electronic signatures in the life sciences and healthcare industries. It ensures the authenticity, integrity, and confidentiality of electronic records, making it indispensable for organizations handling sensitive medical data or developing pharmaceutical products.
A key challenge in achieving 21 CFR Part 11 compliance within the Power Platform environment is the authentication process. While Power Platform typically utilizes silent Single Sign-On (SSO), compliant workflows demand explicit user authentication, including username, password, and Multi-Factor Authentication (MFA).
To address this, we've developed an innovative custom component that initiates the Microsoft login page, enabling full user authentication with MFA without disrupting current login sessions. This solution bridges the gap between Power Platform's standard authentication and the rigorous requirements of 21 CFR Part 11.
By implementing these FDA-compliant workflows, organizations can ensure regulatory adherence, mitigate risks, and maintain the highest standards of data integrity and security in their digital processes.

Some key takeaways include:

1. The solution ensures compliance with 21 CFR Part 11 while leveraging existing Microsoft investments, providing a cost-effective approach to meeting regulatory requirements.

2. It offers enhanced data integrity and security measures, crucial for organizations handling sensitive medical or pharmaceutical data.

3. The streamlined operations and reduced human error can lead to faster time-to-market for regulated products, providing a significant competitive advantage.

Required : Azure AD (Entra access), Power Platform, VSCode, MSBuild, Power Platform CLI components.

Zero to Hero on Azure Functions Security

This session will help you to understand how developers can leverage Azure security features to build a secured Azure Functions app with a live demo application.

In the current digital world, security and privacy are essential. Mostly the data transferred via REST API and developing REST APIs are simplified with the help of Azure Functions. We need to ensure the data transmitted to the outside is secured & followed governed way by the caller rather than anonymous users. This is where we need to implement security on Azure Functions and ensure the Azure Functions are called by authenticated users. In this session, we will dive into the Azure Functions security options, harden your Azure Functions via Azure API & developer best practices.

Create a custom PDF Connectors with Azure Functions and Power Platform: It's Easier Than You Think!

Remote work and automation technology are transforming the way we work, especially in the area of workflow approval processes. These processes are essential in all organizations of all sizes, and automating them can reduce reliance on paper-based approvals and improve efficiency.

Generating PDF documents as part of the audit trail is important during the approval process. This can be done using Adobe Acrobat or other PDF tools like Plumsail, but these tools can be expensive.

Azure Functions is a quick and cost-effective way to generate PDF documents on demand using open-source PDF toolkits like PDFTK. Azure Functions can be exposed as an API, which allows developers to create custom PDF connectors that can be consumed in Power Platform and Logic Apps.

In this session, we will develop a live PDF connector and consume it in Power Platform and Logic Apps.

What, why and how with Azure Chaos Studio !

I am proud to introduce Azure Chaos Studio, the first service of its kind that helps you to perform fault injection on your azure resources (we are talking about network, storage, compute, deployment…) in a safe and controlled way. Imagine the ability to control which faults happen how often and which resources are affected. That's what Azure Chaos Studio does!

You can use Azure Chaos Studio to find and fix potential issues in your Azure environment by injecting controlled faults into your resource, such as disabling a virtual machine or deleting a website. In this way, you can identify weaknesses before the real crisis happens.

Azure Chaos Studio provides a complete automated framework with pre-built functionalities to orchestrate controlled faults, analyse the resilience of your resources, and detect if your deployment is meeting its service level commitments.

In this session, I will show you how Azure Chaos Studio lets you better understand the resiliency of your cloud solutions by balancing risk with reward. You will learn about the range of benefits that Azure Portal can provide when performing chaos experiments.

Twitter Sentiment Analysis using Azure Stream Analytics

In this session, you will learn about sentiment analysis, a new market research method that relies on tweets to understand people's feelings, emotions and attitudes towards a product or brand using Azure Stream Analytics.

Sentiment Analysis is widely used for analyzing opinions, views and conversations on particular topics. Use Twitter Sentiment Analysis for understanding insights from different tweets and get qualitative data. Get up-to-date information on sentiments and trends to make better decisions.

Implementing Dependency Injection (DI) on Azure Functions

Azure Functions supports the dependency injection (DI) software design pattern, which is a technique to achieve Inversion of Control (IoC) between classes and their dependencies. Dependency injection in Azure Functions is built on the .NET Core Dependency Injection features.

In today's session, we will build a simple Azure Function with Dependency Injection.

What is Jupyter Notebooks and why I should care about it?

Jupyter Notebooks is a open source tool and heavily used to develop and execute ML code in the browser or VS code(via extension). Any Jupyter Notebooks document can contain text, images, computational inputs and outputs. In this we will see, how Jupyter Notebooks helps ML developer in Azure Machine Learning environments.

What Azure Data Studio, notebook SQL and why I should care about it?

What is Azure Data Studio,SQL Notebook and why I should care about it?
On November 15, 2017 Microsoft launched a new tool called Azure Data Studio to run on-demand SQL queries, view and save results as text, JSON, or Excel.
ADS (Azure Data Studio) is a cross-platform database tool to manage on-premises and cloud data platforms on Windows, MacOS, and Linux.

ADS provides rich, modern editor experience with IntelliSense, code snippets, source control integration, and an integrated terminal.
It designed mainly for data platform users in mind and it has built in visualization of query result sets.
ADS is not a replacement for SSMS and both has different features and options.
Lets quickly compare the features of SSMS and ADS.

What is Jupyter notebooks and what is the use of it?
What is notebook SQL and it useful in Azure Data Studio?

Introduction and Hands on experience with Azure Stream Analytics

Azure Stream Analytics is a real-time analytics and complex event-processing engine that is designed to analyze and process high volumes of fast streaming data from multiple sources simultaneously.
Patterns and relationships can be identified in information extracted from a number of input sources including devices, sensors, clickstreams, social media feeds, and applications.
These patterns can be used to trigger actions and initiate workflows such as creating alerts, feeding information to a reporting tool, or storing transformed data for later use.
Also, Stream Analytics is available on Azure IoT Edge runtime, and supports the same exact language or syntax as cloud.

There are multiple ways where you can create and configure Azure Stream Analytics job.
* Create a job - Azure portal
* Create a job - Visual Studio
* Create a job - Visual Studio Code
* Create a job - PowerShell

In this session, let's create and configure Azure Stream Analytics job using Azure Portal and Visual Studio.

Develop Azure Dashboards using Azure Monitor and KQL

Azure Monitor (AM) helps us to find the availability and the performance of our applications and services. Using Azure monitor we can collect, analyze and act on telemetry on both cloud and on-premises environments. Azure monitor collects the data in two different ways, metrics and logs. Metric data can be explored via Metric explorer and log data can be queried via Kusto Query Language (KQL). In this session we will see, how we can collect the data, analyze the data using KQL and develop insights and pin them into Azure Dashboard.

Hands on with No-code Automated Machine Learning

Automated Machine Learning(AutoML) is the process of automating time taking / consuming repeated tasks on ML development. This allows the developer to save their time and increases the productivity and quality of the ML models. In this session, I will demonstrate how AutoML helps for AI developers to build their model easily and quickly.

Natural Language to SQL: Building a Conversational Database Interface

In today's data-driven landscape, extracting meaningful insights from databases continues to be a complex challenge. This talk will illustrate how to bridge the gap between human language and database queries can be bridged using Langchain, Python, and SQL. By harnessing the power of natural language processing, we will explore techniques to transform everyday language into precise SQL statements, allowing users to interact with their data more intuitively and efficiently.

We will examine critical aspects such as security, quality and performance to ensure responsible and reliable interactions with natural language database interactions.

Additionally, we'll discuss how these principles can be applied in SaaS applications and extended to visualizations. This session will equip you with a hands-on understanding of building a natural language interface for your database.

Through a live demo, you will witness how to implement these concepts can be put into practice.
By attending, you will gain valuable insights into enhancing user experience, improving data accessibility and addressing security concerns in your data-driven applications.

Stop Worrying, Start Wiring: Azure Serverless for the AI Era

AI is no longer just a buzzword — it’s a fundamental part of modern software. But for developers, the challenge isn’t what AI can do — it’s how to integrate it into scalable, cost-efficient, and maintainable applications. That’s where Azure Serverless comes in.

In this session, we’ll bridge the gap between AI and application engineering. You’ll discover how Azure Functions, Event Grid, and Container Apps can serve as the event-driven backbone of intelligent systems. We’ll explore real-world patterns like automated inference pipelines, orchestrating multi-step AI workflows, and using serverless triggers to activate model predictions or data processing jobs.

To bring it all together, we’ll build and run a live demo: the AI Insights Pipeline — a simple but powerful example where uploading a text file to Blob Storage triggers an Azure Function, calls Azure OpenAI to generate a summary, and sends the result to Microsoft Teams via Logic Apps.

By the end, you’ll understand how to confidently build and deploy intelligent, event-driven solutions with Azure Serverless — without managing infrastructure, worrying about scale, or breaking your budget.

You’ll stop worrying about AI — and start wiring it into everything you build.

After attending this session, participants will be able to:

Identify when and how to use Azure Serverless components (Functions, Logic Apps, Event Grid, Container Apps) for AI-driven workflows.

Build event-driven AI pipelines that react to real-world triggers (file uploads, API calls, or messages).

Integrate Azure OpenAI, Cognitive Services, or custom ML models seamlessly using serverless triggers.

Apply best practices for scaling, cost optimization, and observability for AI workloads.

Design maintainable and secure AI-powered serverless applications.

AI is transforming how developers build and deploy modern applications — but there’s no reason to worry. In this session, we’ll explore how Azure Serverless (Functions, Event Grid, Logic Apps, and Container Apps) helps you integrate AI effortlessly into event-driven, scalable solutions. Through a hands-on demo — the AI Insights Pipeline, where file uploads trigger automatic summarization using Azure OpenAI — you’ll see how to connect intelligence, automation, and cost efficiency. Learn how to wire AI into your apps the right way and take full advantage of the serverless revolution.

Let Your AI Do the Drawing: Agentic Infra Diagrams on Microsoft Stack

Infrastructure diagrams are essential but tedious to maintain. Imagine an agentic AI that listens to your description (“.NET app on a Windows VM connecting to Azure SQL”) and autonomously produces an editable architecture diagram – not just a static image, but a fully editable .drawio file you can refine.

In this session, you’ll build exactly that. We’ll move beyond simple chatbots and create an autonomous agent powered by the Microsoft Agent Framework (in .NET). The agent uses Azure OpenAI (or local LLMs) to understand intent, then orchestrates a series of actions: it calls a Model Context Protocol (MCP) server to generate a Draw.io diagram, and can even open the editor for further tweaks. You’ll see how agentic AI plans, uses tools, and iterates – all locally first to save cloud costs, then deployed to Azure AI Foundry.

Tools & Technologies:

.NET 8 / C# – Microsoft Agent Framework (for building autonomous agents)
Azure OpenAI (or local alternatives like Ollama) – natural language understanding
Model Context Protocol (MCP) – connecting to a Draw.io diagram server as a tool
Draw.io – open‑source diagram editor (embedded or self‑hosted)
Docker – local MCP server containerisation
Azure AI Foundry – production deployment and monitoring

What You’ll Learn:
How to build an agentic AI that plans and uses tools, not just responds
Integrating LLMs with the Microsoft Agent Framework for intent extraction and multi‑step reasoning
Using MCP servers to extend agent capabilities (diagram generation, file I/O)
Generating editable .drawio files and embedding the Draw.io editor
A local‑first development approach (Ollama/LM Studio + Docker) to iterate without cloud spend
The seamless path from local prototype to production on Azure AI Foundry

Audience: .NET developers, cloud architects, and AI enthusiasts eager to move from chatbots to real‑world agentic automation. Basic C# and Azure knowledge helpful but not required.