Luke Edward
Software Engineer at Microsoft
I’m a software engineer at Microsoft in Seattle, where I build LLM tooling and the data infrastructure underneath it. Most of that work is making a model’s answers trustworthy enough that people act on them.
Outside work, I contribute to vLLM-Metal, the Apple Silicon plugin for vLLM, and build WagRead, an iOS app that reads a dog’s mood from a photo. I also play rugby and soccer, surf, and make films.
Work
-
2025 – Now
Software Engineer at Microsoft
Built a natural-language analytics tool on Azure OpenAI that turns questions into validated KQL over telemetry from 2B+ devices. About 100 engineers on 10 teams use it, and it cut their time-to-answer by up to 90%. Also cut a 100TB+ telemetry migration from 13 hours to about 20 minutes, and raised shipping-binary test coverage from 26.4% to over 60%.
-
2024
Software Engineer Intern at Microsoft
Quantified a one-hour battery-life gap across device types, and delivered a two-minute battery gain in Microsoft Defender, plus four minutes on enterprise configurations.
-
2023 – 2024
Software Engineer Intern at Dayforce
Increased API throughput 400% under stress and 630% under load by removing redundant validation calls, and raised API test coverage from 40% to about 85%.
-
2023
Product Research, Design & Development Intern at Interac
Proposed and led a Dockerized escrow REST API (Node.js, Express, PostgreSQL) for a Nest.js and Angular app.
-
Earlier
Founder of BookSmart
Built a salon-booking app and signed 15 salons. Secured grants from IBM and Ontario Centres of Excellence, and won Ontario’s Young Entrepreneurs, Make Your Pitch competition.
Software Engineer Intern at Rogers 5G Create Lab
Built a smart-parking app on 5G sensors for a university smart-city project.
Design Thinking Engineer at RBC
Led a team of four on a React carbon-tracking app and presented it to the C-suite.
Projects
-
2026
vLLM-Metal
Added image input for Gemma 4 after tracking down a per-layer embedding bug that silently corrupted its outputs. A follow-up change cut image time-to-first-token by about 30% and startup memory overhead by 25% by computing logits only for the rows the sampler reads.
-
2025 – Now
WagRead
An iOS app I built and run on my own. Gemini reads a dog’s mood from one photo and pins each sign it sees on the dog; a separate health check covers coat, eyes, skin, weight and teeth. Rebuilt natively in SwiftUI in 2026, with model calls hedged across two models instead of a serial fallback that could take 30 seconds.
Education
-
2025
BSc in Computer Science at Western University
GPA 3.8 and Dean’s Honor List. RBC Data Science Scholarship, Ted Rogers Scholarship and two undergraduate research grants.
Contact
-
Email
-
GitHub
-
LinkedIn