WOW-FACTOR OF THIS PROJECT /
Turning Accessibility into Universal Performance: I translated neurodivergent workflow user research into 3 AI-powered features that optimized productivity for all professionals.
System-led Design: I architected a dual-layer experience with intuitive UI on top and a research-grounded Custom AI-System Instruction as a backend, ensuring every AI output felt user-aligned.
PROBLEM /
At launch in 2024, Azure AI Foundry had strong technical foundations but lacked real-world, inclusive use cases that demonstrated its productivity value across Microsoft 365. Without ROI-driven stories, enterprise adoption remained slow.
MY ROLE & OUTCOME /
As a Product Designer and Technical Lead, I led the creation of Flow, a concept 0–1 desktop app with three neurodivergent-inspired AI features and a custom AI system instruction that optimized workflow.
IMPACT /
Flow saved professionals an avg of 25 mins/day, achieved 100% pilot adoption, and scaled accessibility features into universal workflow solutions that validated Azure AI Foundry’s value across Microsoft 365.
RESPONSIBILITIES /
User research, End-to-End Prototyping, Stakeholder Alignment, Usability Testing, Technical lead (UX/AI Engineering), Prompt Engineering & Evaluation
PROJECT TYPE /
Corporate Sponsor
TIMELINE & OUTPUT /
9 months (Dec 2024 – Aug 2025) to deliver 0–1 desktop app + custom AI-Assistant
TEAM /
1 Principal PM, Microsoft
1 Senior UX Researcher, Microsoft
4 Product Designers
SPONSOR COMPANY /
Microsoft, Azure AI Foundry
+25 mins/day
Saved per user/day (5 users)
+40%
Clarity improvement
in AI outputs (25 users)
100%
Pilot Adoption Rate (25 users)
Context
The Gap: Turning Azure AI’s Technical Strength into Proof of Enterprise Value
What is Azure AI Foundry
Launched in 2024, Azure AI Foundry is Microsoft’s platform for building and managing generative AI apps and agents across the enterprise ecosystem.
Microsoft Azure AI's Challenge
Azure AI had strong technical capabilities but lacked inclusive, ROI-driven use cases that proved productivity impact that stalled enterprise adoption.
Project Goal
Translate Azure AI Foundry’s capabilities into neurodivergent-inspired, Microsoft 365-ready features that demonstrate real business value. By turning accessibility into a productivity advantage, I aimed to de-risk enterprise adoption and move Azure AI from pilot to scale.
Target User Group & Why Neurodivergence Focus
The target users were working professionals. Starting with neurodivergent insights revealed universal workflow pain points that affect all professionals.
Defining Problem
Universal Daily Pain-Points for Professionals:
Communication Block, Context Switching & Scattered Focus
To uncover everyday workflow pain points, I led a survey of 45 neurodivergent professionals and in-depth interviews with 5 users and SME (Clinician, ADHD Coach, ADHD-productivity tool startup).
Click to see the final Research Report
User Survey (45 user responses)
Semi-Structured Interviews with Users & SMEs (7 in-total)
IDENTIFIED PAIN-POINTS
By researching neurodivergent workflows, I identified three universal productivity barriers: constant context switching, communication block, and scattered focus. These insights reframed accessibility as a driver of universality, not a niche constraint.
PAIN-POINT 01.
Communication Block
50%
Feel negatively affects performance
PAIN-POINT 02.
Context Switching
275 pings/day
avg interruptions in one's workflow
PAIN-POINT 03.
Scattered Focus
40%
Feel negatively affects productivity
AZURE AI CAPABILITIES LEVERAGED
To translate research into action, I mapped each pain point to Azure AI Foundry’s core technical strengths, identifying where AI could best intervene.
Click to see the Azure Cloud & Azure AI Foundry tech guide I created for the team
By bridging user insight with technical capability, I demonstrated how inclusive design can scale productivity for all professionals. These findings became the blueprint that turned Azure AI’s potential into proof, shaping Flow’s design strategy and system architecture.
Final Design
Flow: Turning Accessibility into Enterprise Productivity (Three AI-powered Features + Custom AI-System Instruction Layered)
Flow is a context-aware desktop app designed for the Microsoft 365 ecosystem.
It delivers three AI-powered features (Magic Mouse, Activity History, Context Bookmark) all powered by a Custom AI-System Instruction built on Azure AI Foundry that reduces cognitive load and enhances workflow optimization for professionals.
Flow - active state on a desktop
Zoom-in: Flow's Three Key Features
A Custom AI-System Instruction powers each feature, translating cognitive research into structured logic that drives clarity and focus.
Built and deployed on Azure AI Foundry, this layer converts empathy-only responses into role-aware, actionable guidance that scales across Microsoft 365.
Click to see the final Custom AI-System Instruction
Key Feature 01
Magic Mouse:
Captures on-screen context and sends it directly to the LLM, removing the need to explain from scratch
PAIN-POINT: Cognitive Load & Communication Block
Ugh - I need to write the whole context from SCRATCH on my LLM
SOLUTION
One quick selection/highlight of an area instantly becomes an input for my LLM!
IMPACT
45-60sec
Saved/ prompt writing
Switch between Text & Area Select Mode
Interact with a Custom AI Assistant
Key Feature 02
Activity History:
Instantly regain context & resume task through AI-powered searchable timeline
PAIN-POINT: Context Switch
I got interrupted again. What was I working on?
SOLUTION
With one Copilot command on this timeline, I can instantly retrace my steps!
IMPACT
3-4 sec
Estimated Time that takes to return to the previous task
IMPACT
Proving Azure AI’s Value through Inclusive AI Design
By transforming accessibility into a mainstream productivity driver, Flow turned Azure AI Foundry’s technical promise into enterprise adoption and measurable business proof.
🔹 Business Impact
Business ROI Projection
+457%
ROI with Flow integrated into Microsoft Copilot
Productivity Saved
+25 mins/day
Time-Saving per user from productivity boost using Flow
🔹 User Impact
Adoption Usability
100%
Pilot adoption intent
AI Outputs
+40%
Clarity Improvements
🔹 Stakeholder Approved
“You should patent or publish paper on this. I’ve never seen anything like what you shared.”
— Microsoft Principal PM, Azure AI
⭐⭐⭐⭐⭐
“I think [the solution] resonates really well. We all need help for problems like this."
— Microsoft Senior Product Designer
⭐⭐⭐⭐⭐
Design Process
Design Decision for a Dual-layer Solution:
UI features + Custom AI-System Instruction
WHY DUAL-LAYER?
I led the decision to design a dual-layer architecture because some pain points required a system-level layer to ensure inclusive AI outputs from backend; others needed UI-level tools for direct user control.
Together, these layers connected how AI thinks with how professionals work, ensuring both intelligence and agency.
VISIBLE LAYER
UI-Layer
Phase 1. Feature Exploration 0-1
INITIAL APPROACH
I led an exploration design workshop and synthesis process using sticky-note ideation grounded in four research-driven design principles.
Design Workshop: From Design Principle-driven Ideation -> Synthesizing into themes
DESIGN PRINCIPLES
01
Be quietly available, never intrusive
Assist only when needed, stay invisible when not.
02
Assist me like you know me
Learn patterns, adapt to attention cycles and habits.
03
Let me practice my thoughts without pressure
Offer judgment-free, context-specific help.
04
Make engagement purposeful, not just playful
Spark motivation with meaningful, novel cues.
RESULT
Through affinitization, I synthesized over 50+ ideas into seven potential feature directions below.
Initial Design (mid-fidelity): seven potential feature directions
Phase 2. Core Feature Prioritization
CONCEPT TEST
To validate direction before investing in build-level fidelity, I facilitated five 60-minute concept-testing sessions with senior Microsoft designers and professionals. The goal was to evaluate which ideas had the strongest potential for adoption.
What I learned
Too many features diluted focus
Simplicity drives adoption
Universal pain points carried more value than niche ones
From this, I created three scoping criteria that aligned design focus with business feasibility and product vision:
FURTHER SCOPING CRITERIA
01
Solve the Right Problem
Does it directly address core pain points?
02
Leverage Azure AI Strengths
Does it map to Azure AI Foundry’s core capabilities?
03
Real Use-Case
Will professionals see immediate value inside Microsoft 365?
RESULT
This process and criteria guided us to scope down to three core features that best balanced user needs, technical feasibility, and product vision. This became the foundation for Flow’s dual-layer architecture, merging user-facing tools with a custom AI system instruction.
Final Design (high-fidelity): three core feature directions
Phase 3. Core Feature-level Iteration: Activity History
USABILITY TEST
After down-selecting the core features, I conducted 4 usability tests (60 mins each) with professionals on Magic Mouse, Activity History, and Context Bookmark. Feedback led to refinements, and I owned the Activity History iteration.
ITERATION: ACTIVITY HISTORY
Activity History passively captures your on-screen activity and turns it into an AI-powered timeline, so you can instantly retrace what you were doing and why without losing context.
The initial Activity History relied on manual timeline navigation, yet usability testing revealed two critical insights for refinement:
INSIGHT #1 - Interaction Mode
3 of 4 users preferred Copilot chat interaction over manual timeline navigation.
REFINEMENT
Made Copilot the default entry point for Activity History.
INSIGHT #2 - LLM Output
Users wanted to know not just what they did, but why it mattered.
REFINEMENT
Redesigned the summary output to surface the “why” behind each activity.
RESULT
These refinements not only improved of giving higher clarity and faster recall, but also strengthened Copilot’s role as the central interaction model, directly supporting Microsoft’s product vision of driving Copilot adoption across the ecosystem.
INVISIBLE LAYER
AI System Layer
Problem with the Default LLM (Copilot)
WHY DEFAULT LLM (Copilot) WASN'T ENOUGH
I found that default Copilot outputs were empathetic yet generic and unstructured, creating cognitive overload that conflicted with the project goal of optimizing daily workflows.
Tone: Generic
Unstructured
Risk of Cognitive Overload
SOLUTION: CUSTOM AI-SYSTEM INSTRUCTION
I designed a Custom AI-System Instruction to deliver concise, structured, role-specific guidance grounded in user research. By shifting from empathy to action, it turned Copilot from a “helpful voice” into a practical workflow partner.
Tone: Grounded
Clear Steps
4.3/5⭐User Satisfaction
Click to see the final Custom AI-System Instruction
Comparison between the default LLM v. Flow's custom-AI Output
Iteration: Evaluating AI Prompts Beyond Accuracy (3 Rounds)
IMPORTANCE & MY ROLE
Most AI prompt evaluations stop at technical performance for accuracy, but those don’t guarantee usefulness or adoption.
I developed a 3-layer evaluation framework by combining internal heuristic testing, SME & user-based scenario validation, and Azure AI Foundry’s quantitative scoring. This ensured the model was not only technically sound but adoption-ready.
OUTCOME
The custom AI-system layer turned Flow from a concept into a proof of adoption. By uniting cognitive insight with technical precision, Flow demonstrated how inclusive intelligence can scale reliably across enterprise systems.
Reflection
From Systems Thinking to What’s Next
WHAT I LEARNED
This project taught me that designing for AI is building connected systems where UX, data, and engineering work as one.
I learned to translate technical feasibility, user needs, and business & product priorities into design outcomes that scale trust and adoption.
By combining cognitive research with system-level thinking, I created a dual-layer architecture solution that bridged how AI thinks with how users work, ensuring clarity and reliability at scale.
IF I COULD DO IT AGAIN
I’d explore a multi-agent orchestration layer to evolve Flow from an individual assistant into a coordinated AI ecosystem. Through specialized agents, Flow could manage cross-functional workflows, integrate with enterprise pipelines, and drive distributed intelligence across Microsoft 365.
Multi-Agent Orchestration Architecture of Flow
(evolving from One supportive AI -> Specialized Agents)
Sample Demo of Multi-Agent Orchestration I tried on Azure AI Foundry