Microsoft Internship

Microsoft

Tutorial & Documentation

Accessibility compliance manager

UX Copywriter

AI/Machine Learning + Edge Computing for Retail & Open Source Tutorial

Empowering Edge Intelligence

Projects completed by the inaugural Microsoft Garage Silicon Valley Intern Team, Summer 2019

View more working details

Project Type

Open Source Tutorial & a Retail Solution

Duration

12 weeks, Summer 2019

Tools

Microsoft Office Suite

Team

1 UX Designer and 5 Software Developers

Non Disclosure Agreement

My team's primary challenge was publishing an open source tutorial on how to integrate Azure with machine learning on the Jetson Nano (an ARM64 device). Our second challenge was to develop an enterprise retail solution which is still under a non disclosure agreement. This portion of the project has not been included in this summary page.

The team from left to right: Abhinav Ayalur, Kaden Dippe, Lindsey Cleary, Angela Martin, Kelly Lin, and Priscilla Lui

Non Disclosure Agreement

The team from left to right: Abhinav Ayalur, Kaden Dippe, Lindsey Cleary, Angela Martin, Kelly Lin, and Priscilla Lui

Design Digest

As the primary UX writer and user tester on the inaugural Microsoft Garage intern team in Silicon Valley, I collaborated with a team of four intern software developers and partnered closely with stakeholders from Azure Machine Learning, ONNX Runtime, and Azure Data Box Edge. Our goal was to help bring machine learning to the edge by making highly technical workflows more approachable and accessible for developers working with constrained hardware.

I co-authored an open-source tutorial for deploying ONNX models on the NVIDIA Jetson Nano using Azure IoT Edge. My focus was on rewriting complex technical documentation with clarity, inclusivity, and usability in mind—ensuring developers of all backgrounds could implement edge AI solutions with confidence. Our work contributed to the ONNX Runtime 0.5 release, which introduced support for hardware acceleration at the edge, and the tutorial is now part of Microsoft’s official tooling to help democratize AI and machine learning at scale.

Integrate Azure with Machine Learning on the Jetson Nano

The Microsoft Garage intern program brought together a cross-functional team to explore practical applications of AI and machine learning at the edge. Our challenge: to create a developer-friendly, open source solution that integrates Azure Machine Learning with edge devices, specifically the NVIDIA Jetson Nano.

Integrate Azure with machine learning execution on the NVIDIA Jetson platform (an ARM64 device)

In this tutorial you will learn how to integrate Azure services with machine learning on the NVIDIA Jetson device (an ARM64 hardware platform) using Python. By the end of this sample, you will have a low-cost DIY solution for object detection within a space and a unique understanding of integrating ARM64 platform with Azure IoT services and machine learning."

View our open source tutorial

Read our press release article

Discover

We researched barriers developers face when deploying machine learning models on edge hardware—particularly ARM64-based devices. Through stakeholder interviews and reviews of technical documentation, we uncovered that setup complexity, hardware-specific quirks, and lack of streamlined tutorials were major blockers.

Problem Statement

AI developers and IoT engineers face significant challenges when deploying machine learning models to edge devices, particularly ARM64 platforms like the NVIDIA Jetson Nano. These challenges include complex setup processes, inconsistent documentation, and limited support for hardware acceleration.

Problem Statement

Vision Statement

To empower developers and researchers to easily deploy high-performance machine learning models to edge devices—unlocking the full potential of AI at the edge through accessible, well-documented, and hardware-accelerated workflows.

Vision Statement

Define

We identified an opportunity to create a modular, open source tutorial that demystifies edge AI deployment. The target audience was AI developers and IoT engineers looking to accelerate inference on lightweight, GPU-enabled edge devices using familiar Microsoft tools.

Project Goals

1) Lower the barrier to entry for deploying ML at the edge 2) Promote ONNX Runtime and Azure ML as a viable edge inference stack 3) Demonstrate Microsoft's commitment to open-source and accessible AI

Project Goals

1) Lower the barrier to entry for deploying ML at the edge 2) Promote ONNX Runtime and Azure ML as a viable edge inference stack 3) Demonstrate Microsoft's commitment to open-source and accessible AI

Project Goals

1) Lower the barrier to entry for deploying ML at the edge 2) Promote ONNX Runtime and Azure ML as a viable edge inference stack 3) Demonstrate Microsoft's commitment to open-source and accessible AI

Proposed Solution

There is a need for a streamlined, accessible, and modular tutorial that demonstrates how to integrate Azure Machine Learning with ONNX Runtime to deploy models efficiently on GPU-enabled edge hardware.

Proposed Solution

Challenges

1) ARM64 device limitations (Jetson Nano’s power, compute, OS), which made the tutorial necessary in the first place 2) Required integration with existing Microsoft tools (Azure ML, IoT Edge, ONNX Runtime) 3) Tight internship timeline and coordination with multiple internal teams

Challenges

Design

As the UX designer, the deliverables for this internship project were a little different than I had previously contributed to. It was exciting to have led the design of the tutorial structure and supporting assets as a content strategist and copywriter. I mapped the process, optimized developer experience through UX writing, and worked closely with PMs and engineers to ensure clarity, accessibility, and technical accuracy. I also advocated for inclusive documentation practices to support diverse developer backgrounds.

Solution: Tutorial Design & Documentation

We created an open-source tutorial and reference implementation that demonstrates how to integrate Azure Machine Learning with ONNX Runtime on the NVIDIA Jetson Nano. The solution simplifies deployment by providing containerized, modular code and end-to-end guidance—from model training in the cloud to accelerated inference at the edge—enabling developers to focus on innovation rather than infrastructure.

Solution: Tutorial Design & Documentation

Collaborative Drafting Process

Writing the tutorial was a highly collaborative, iterative process that required both technical accuracy and user-centered clarity. As the sole UX designer on a multidisciplinary team, I worked closely with our four software development interns to translate our working prototype and deployment pipeline into a step-by-step guide that developers could easily follow.

Collaborative Drafting Process

To ensure technical precision, we consulted regularly with engineers and program managers from our stakeholders on the Azure Machine Learning, ONNX Runtime, and Azure Data Box Edge teams. These experts helped validate our instructions, confirmed compatibility across platforms, and advised on best practices for deploying ONNX models using Azure tools.

I led the drafting and UX writing of the tutorial, focusing on:

Clear labeling and structure of the steps (e.g., container setup, model export, edge deployment)
Explaining complex workflows using plain language
Anticipating developer pain points and surfacing troubleshooting tips

User Testing

Each draft of the tutorial went through multiple rounds of review—first with our internal team, then with external stakeholders—to align both technical requirements and user experience. To validate the clarity, usability, and technical accuracy of our tutorial, we conducted live user testing during two high-impact events: our team’s Demo Day presentation and the 2019 Microsoft Hackathon showroom floor.

User Testing

Demo Day

At our Demo Day, we had the opportunity to present our work and introduced the tutorial to an audience of Microsoft employees, interns, and product leaders. As the UX designer, I observed how attendees interacted with the documentation in real-time—where they paused, what questions they asked, and whether the instructions led to successful deployment on the Jetson Nano. These live interactions helped us identify key areas where we needed to clarify steps, simplify language, or better explain dependencies. Our presentation and demo day was such a success, we were featured on Microsoft's internal news page as the top story!

Demo Day

Hackathon 2019

Later, at the Microsoft 2019 Hackathon showroom floor, we tested the tutorial with a broader range of engineers, developers, and AI enthusiasts.

Hackathon 2019

Later, at the Microsoft 2019 Hackathon showroom floor, we tested the tutorial with a broader range of engineers, developers, and AI enthusiasts.

Hackathon 2019

Later, at the Microsoft 2019 Hackathon showroom floor, we tested the tutorial with a broader range of engineers, developers, and AI enthusiasts.

This setting allowed us to:

Watch first-time users navigate the tutorial end-to-end

Get feedback from developers with varied levels of familiarity with ONNX, Azure IoT Edge, and embedded devices
Validate that the tutorial performed reliably across multiple hardware setups

Insights & Iterations

From these sessions, we made UX-informed refinements such as reordering key setup instructions, embedding more direct links to Azure tools, and adding diagrams and photos to reduce cognitive load. These changes ultimately made the tutorial more accessible, repeatable, and scalable—contributing to its publication as an official Microsoft open-source resource.

Insights & Iterations

Deliver

We released the open source tutorial on GitHub: ONNX Runtime IoT Edge with Azure and Jetson Nano. The solution uses Dockerized containers to streamline deployment, leverages ONNX Runtime with TensorRT for GPU acceleration, and captures telemetry for remote monitoring via Azure.

Now Live: Open-Source Tutorial for Deploying AI to the Edge

Created in collaboration with the Azure Machine Learning, ONNX Runtime, and Azure Data Box Edge teams, the tutorial was designed to make powerful AI and machine learning more accessible—especially on low-power, resource-constrained devices. Now featured as part of the official ONNX Runtime 0.5 release, this guide helps developers streamline their edge inferencing workflows with clarity, usability, and inclusion at its core. Now available on GitHub!

Now Live: Open-Source Tutorial for Deploying AI to the Edge

View our open source tutorial

Featured on Microsoft Open Source blog

Our project was spotlighted in a featured article on Microsoft’s Open Source blog, showcasing our work on ONNX Runtime 0.5 and its support for edge hardware acceleration. The post highlights the tutorial we co-developed for deploying ONNX models to the NVIDIA Jetson Nano—demonstrating how our open-source solution helps make AI at the edge more accessible and powerful for developers around the world.

Featured on Microsoft Open Source blog

Read our press release article

Showcasing Our AI Edge Tutorial on Microsoft’s Platform

At the conclusion of our 12-week internship, I had the opportunity to present our project outcomes to collaborators from Azure Machine Learning, ONNX Runtime, Azure Data Box Edge, and the Microsoft Garage leadership team.

Showcasing Our AI Edge Tutorial on Microsoft’s Platform

Debrief

Our open-source tutorial was featured in the ONNX Runtime 0.5 release and highlighted in a Microsoft Open Source spotlight, demonstrating a commitment to making edge AI more accessible. Validated on the NVIDIA Jetson Nano, the tutorial now serves as a reference implementation for deploying machine learning models on IoT devices. This cross-functional effort underscored the importance of collaboration, inclusive content strategy, and user-centered design in developer tooling.

During my internship, I grew as a UX designer, accessibility advocate, and content strategist. I learned how to communicate across technical and non-technical roles, design for a variety of learning styles, and champion clarity in complex technical documentation. Most importantly, I saw firsthand how vital user testing is—helping us uncover blind spots and improve the experience in ways we couldn’t have anticipated on our own.