Microsoft
Azure Internal Tool
Department:
Azure Governance
Timeframe
2023.5 - 2023.12
Role
Lead product designer
This internal-facing initiative focused on improving productivity and decision-making for Azure engineering and PM teams by delivering two complementary tools:
1. Azure Terraform Resource Coverage Portal
A centralized dashboard that helps PMs and resource teams track the completeness of Terraform support across Azure services. The portal provides multi-level filtering, ownership clarity, and exportable insights to prioritize engineering effort and reduce duplication across teams.
2. ARM AI Assistant
An AI-powered assistant integrated into the Azure Resource Manager ecosystem, designed to help internal users explore API specifications, query schema details, and troubleshoot deployment issues using natural language. The assistant reduces ramp-up time and enhances discoverability for both new and experienced engineers.
Together, these tools support internal alignment, improve resource planning, and accelerate platform development at scale
Design overview
01 - Impact at a Glance (Expected)
Around 1500 Internal Users
Resource coverage portal used by Azure PMs, engineers, and platform teams across multiple orgs.
23% increase in user confidence
Resource coverage portal reported greater clarity when planning Terraform priorities.
1000+ Internal Users
Used by engineers and internal support teams to explore ARM specs and resolve deployment issues
–35% Reduction in Internal Support Cases
By enabling self-service troubleshooting and API discovery via natural language queries
Design overview
02 - Project Snapshot
Azure Terraform Resource Coverage Portal
Target user:
Azure product managers and engineering leads responsible for Terraform enablement
User pain point:
No centralized way to track which Azure resources are supported in Terraform, leading to duplicate efforts and unclear ownership
Product goal:
Provide a clear, filterable dashboard to monitor coverage progress, identify gaps, and export insights for planning.
Challenge:
Display large volumes of service data in a way that supports both high-level prioritization and deep drill-downs
ARM AI Assistant
Target user:
Azure engineers, service teams, and support staff working with ARM templates and APIs
User pain point:
ARM schema and deployment issues are difficult to navigate, resulting in high internal case volume and long resolution times.
Product goal:
Enable natural language exploration of ARM specs and self-service troubleshooting through an integrated AI assistant.
Challenge:
Balance AI flexibility with technical accuracy while keeping the experience lightweight and trustworthy for internal users
From user insight to focused execution
Design Process & Key Decision
Design process & key decision
Design Process
Both tools were initiated in response to clear internal needs voiced by Azure product and engineering teams. Rather than emerging from top-down product roadmaps, these projects were driven by real pain points: duplicated engineering efforts in Terraform enablement, and high internal support volume caused by complex ARM schema navigation.
1. Internal Need Alignment
Direct collaboration with internal stakeholders—including PMs, engineers, and support leads—helped document key pain points, workflow inefficiencies, and desired outcomes.
2. MVP Scoping
Core functionality was prioritized to address the most immediate user needs while maintaining implementation feasibility and scalability.
2. High-Fidelity Prototyping
Early design hypotheses were translated into high-fidelity prototypes: a dashboard-driven experience for Terraform visibility and a conversational assistant for self-service ARM support.
3. Targeted Interviews & Feedback Loops
Lightweight user interviews and async reviews helped validate assumptions, clarify terminology, and adjust interaction models—especially around data detail levels and AI response expectations.
Design process & key decision
01 - Internal Need Alignment
Azure Terraform Coverage Portal Internal Need
01/ Lack of layered visibility
Users couldn’t easily filter or segment Terraform support data by team, region, or resource type—making it difficult to prioritize or track progress.
02/ No centralized coverage view
There was no intuitive way to check which Azure resources were supported in Terraform across the platform, leading to scattered, inconsistent tracking.
03/ Repetitive and manual exempt request process
Submitting Terraform exemptions was time-consuming and uncoordinated, often resulting in duplicated requests and unnecessary overhead for PMs and engineers.
ARM AI Assistant Internal Need
01/ High volume of repetitive internal support cases
Support teams were overwhelmed by recurring questions about ARM schema usage, deployment issues, and property configurations.
02/ Lack of self-service troubleshooting tools
Engineers had no simple way to explore ARM specs or resolve issues independently, especially if they were new to the platform.
03/ Overreliance on SMEs for routine tasks
Even simple requests required direct input from subject matter experts, creating bottlenecks and slowing down developer productivity.
Design process & key decision
02 - MVP Scoping
Azure Terraform Coverage Portal Internal Need
01/ Centralized Coverage Overview
Design a single dashboard to visualize which Azure resources are supported in Terraform, improving transparency across teams.
02/ Multi-dimensional Filtering & Ownership View
Enable filtering by team, region, resource type, and service to help users focus on what matters and understand ownership at a glance.
03/ Streamlined Exempt Request Tracking
Consolidate and label duplicate exempt requests to reduce redundancy and improve processing efficiency.
ARM AI Assistant
01/ Natural Language Query Support
Allow users to ask questions about ARM schemas, properties, and errors using plain language to enable self-service learning.
02/ Integration with Internal Knowledge Base
Connect the assistant to a curated set of common case resolutions to reduce repetitive support workload.
03/ Lightweight Integration & Extensibility
Embed the assistant into developer tools or documentation portals with a scalable architecture for future expansion.
Design process & key decision
03 - High-Fidelity Prototyping
Based on the MVP pre defined, I generated the design for Azure Terraform Resource Coverage Portal
Azure Terraform Resource Coverage Portal Design
Scope 1: Centralized Coverage Overview
Scope 2: Multi-dimensional Filtering & Ownership View
Scope 3: Streamlined Exempt Request Tracking
ARM AI Assistant
Scope 1: Natural language query support
Scope 2: Integration with internal knowledge base
Scope 3: Lightweight integration extensibility
Design process & key decision
04 - Targeted Interviews & Feedback Loops
Azure Terraform Resource Coverage Portal
The exempt request (“fill request”) process was identified as a repeated source of friction during usability testing. While users appreciated having a formal way to submit exemption requests for unsupported resources, the experience was described as tedious, repetitive, and overly manual—especially when submitting multiple similar requests. Key issues included lack of pre-filled data, no bulk selection, and minimal context awareness.
Although the design team explored improvements such as smart defaults, batch request support, and validation prompts, these enhancements were deprioritized due to technical complexity and backend governance logic that was still under development. Implementing dynamic validation and automated input suggestions would require deeper integration with internal metadata systems and policy frameworks, which were outside the MVP scope.
The current experience offers a working baseline, but there’s clear opportunity for future optimization once platform maturity and engineering bandwidth allow.
ARM AI Assistant
After several rounds of iteration, the ARM AI Assistant effectively supports basic queries and troubleshooting. However, a recurring user pain point remains: when queries lack clear context or resource type, the assistant struggles to identify the appropriate knowledge base. As a result, users often need to manually switch sources to get accurate answers. While the design considered adding contextual intent detection, current backend limitations—such as the lack of unified tagging and intent parsing—make this technically infeasible for now. This improvement remains on the roadmap for future refinement once platform capabilities mature.
Balancing Vision and Constraints
Design Summary
This project reinforced the belief that internal tools should be treated with the same level of design rigor as customer-facing products. By grounding our solutions in real, day-to-day pain points and working closely with stakeholders across PM and engineering, we delivered tools that not only improved operational efficiency—but also built long-term trust in design as a strategic partner within Azure.
Thoughtful UX, even behind the scenes, can be a powerful force multiplier—especially when it enables thousands of engineers to move faster, make smarter decisions, and focus on what matters most.