Inside this project
Leading Design & Strategy
I led the design evolution from hackathon prototype to core platform feature, shaping the interaction model and visual identity. This work relied on close collaboration with a cross-functional team including:
- Product managers who I collaborated with to define the product vision
- Engineers who built the AI-powered systems
- Content designer who I partnered with to craft the AI personality and responses
Problem & Overview
Microsoft AI Foundry is a developer platform for discovering, configuring, and deploying AI models and tools. As the platform grew, developers increasingly struggled with documentation discovery, complex configuration options, and operational workflows such as managing quotas or deploying models.
Ask AI began as a hackathon project exploring whether an in-product assistant could reduce this friction. The project won the hackathon and received leadership support to continue building it into the product.
How can we help developers understand and operate Foundry without breaking their flow?
An introduction to the Foundry platform.
Hackathon Prototype
Answers & Documentation
An AI-powered chat assistant embedded throughout Azure AI Foundry, helping users discover models, compare options, and get instant answers without leaving their workflow.
Answer Questions
Provide clear explanations about Foundry concepts and workflows
Surface Documentation
Make relevant documentation easily discoverable without leaving the product
Focus on Learning
Explanation rather than action to build trust and understanding
From Explaining to Doing
After the hackathon, the project continued with leadership support. As confidence grew, we expanded Ask AI's capabilities by introducing MCP-based support, allowing it to connect to other agents behind the scenes.
MCP (Model Context Protocol) support connected the assistant to other agents behind the scenes.
New Capabilities Enabled
- Managing quotas
- Deploying models
- Supporting operational workflows tied to Foundry configuration
New UX Challenges
This shift from passive explanation to active execution introduced new design challenges:
- Communicating when the assistant could perform actions vs. provide guidance
- Designing confirmation and feedback for agent-driven operations
- Maintaining trust and clarity in high-impact workflows
Foundry Refresh
Following a major Microsoft AI Foundry refresh, we revisited the assistant's role. Originally, Foundry Agent existed as a discrete assistant experience. With the refresh, we repositioned it as Ask AI, a core interaction model integrated directly into Foundry's primary search experience.
Search-Integrated AI
Ask AI integrated into the primary search bar, making AI assistance the default interaction model
- Ask AI became tied to the main search bar
- Searching, asking questions, and initiating AI-assisted workflows shared a single entry point
- AI assistance became part of how users navigate and operate Foundry
Ask AI integrated with the primary search experience.
Privacy & Security First
With expanded agentic capabilities, we designed robust privacy and security controls from the ground up, ensuring users maintain full control over what actions the AI can perform on their behalf.
Security Features
- Explicit permission controls for each agentic capability
- Clear visibility into what actions AI can and cannot perform
- Ability to revoke permissions and audit AI actions at any time
- Trust and transparency built into every interaction
Inline Agentic Assistance
Exploring ambient AI support within the natural workflow
With Ask AI embedded in the core UX, we are now exploring inline agentic support — a strategic middle ground between traditional UI and conversational interfaces.
The Challenge: Full chat conversations can be tedious for simple tasks, yet traditional UI becomes overwhelming when users face complex decisions with numerous parameters. Inline assistance offers limited but highly contextual functionality — surfacing AI guidance precisely where complexity demands it, without forcing users into a separate conversation.
The goal is to make AI assistance feel ambient and natural, complementing the side-panel assistant with contextual, inline help that appears exactly when and where it's needed—without overwhelming users or breaking their flow.
An exploration of contextual support within the workflow.
Design Influences
Microsoft Copilot Design System
Foundation for conversational UI patterns, tone of voice, and interaction principles
Market Leaders
Studied competitor approaches to AI assistants, chat interfaces, and agentic workflows
Multimodal Chat Playground
Exploration space for various input modalities and rich interaction patterns
Agentic Playgrounds
Testing environments for agent behaviors, tool calling, and workflow orchestration
Foundational systems and research informing the Ask AI experience
The multimodal chat and agentic playgrounds within AI Foundry were also part of my design responsibilities. This allowed me to create a cohesive design language across different product areas while exploring cutting-edge interaction patterns that could inform the Ask AI experience.
From Hackathon to Platform Capability
Hackathon to Production
Successfully transitioned concept from hackathon win into a supported product experience
Evolved Capabilities
Grew from documentation Q&A into a full agent-backed system supporting operational tasks
Core Platform Integration
Became a central part of Foundry's refreshed UX and primary interaction model
Search-Integrated AI
Established patterns for search-integrated AI assistance as default workflow