Experience · Microsoft · April 2021 – Present

From product engineering to the quality of AI systems.

I joined Microsoft in April 2021. My work has moved through three chapters: product engineering on Microsoft Feed, building AI-powered product experiences, and measuring and improving the quality and reliability of AI responses.

Chapter 01 · April 2021 – September 2023

Product and Platform Engineering

Product engineering on Microsoft Feed in Outlook mobile: shipping features, integrating social signals, measuring engagement and owning the experience in production.

Feed product engineering, including SharePoint News Reactions
Social signals and integration work
Engagement measurement and telemetry, including dwell-time tracking
Performance and reliability improvements
Accessibility and production ownership, including on-call support

01 Professional

Building and Improving Production Feed Experiences

Product engineering on Microsoft Feed across web and mobile: shipping features, measuring engagement, and keeping the experience fast, accessible and reliable in production.

Microsoft · 2021–2023Read case study

Chapter 02 · Late 2023 – 2026

AI-Powered Product Experiences

From late 2023 my work moved from conventional product engineering towards AI-powered experiences in Microsoft 365: first features that summarise and surface information inside existing products, then experiences built on agent workflows.

I worked on Copilot in Outlook and Meeting Digest, Catch Up in Microsoft 365 Copilot, the Catchup Agent, and Daily Briefing (Copilot Briefs). Not all of these were released publicly.

The engineering problems changed with them. Instead of rendering a known data model, the client had to work with agent workflows: requesting work, following its state, streaming and rendering results as they arrived, and failing clearly when something went wrong.

Building AI-powered product experiences
Integrating client experiences with agent workflows
Retrieval and grounding
GraphQL integration
Streaming, rendering and reliability
Testing and evaluating AI-powered workflows

Examples from this work

Proposed and implemented error mapping so that clients received meaningful error information instead of a generic code from a backend AI service.
Traced an authentication failure after a workflow migration to the wrong kind of token, and shipped the fixes.
With a colleague, fixed false error messages that appeared even when a completed answer existed.
Helped create initial monitors and troubleshooting guides for common failures.
Consolidated prompt snapshot tests so that every prompt type was covered automatically.
In a design review for moving functionality behind a central gateway, raised questions about retries, stale-data detection and how much pipeline state to expose to clients.

Chapter 03 · Primarily 2025 – 2026

AI Quality, Evaluation and Reliability

Measuring whether AI-generated responses are actually good, finding out why they fall short, and making the evidence behind them more reliable.

Two different engineering problems, each with its own case study: Researcher, on measuring and improving response quality, and Copilot Citations, on the reliability of the sources behind a response.

01 Professional

Researcher: Measuring and Improving AI Response Quality

Evaluation metrics for information density and temporal relevance, offline baselines and experiments, and investigations into why AI-generated reports lose context or fall short.

Microsoft · 2025–2026Read case study 02 Professional

Copilot Citations: Improving Response Reliability

Citation handling and rendering, monitoring and telemetry, and reliability investigations into how sources travel from an AI response to the screen.

Microsoft · 2025–2026Read case study

Now · since September 2026

Microsoft 365 Copilot Tuning

My current area of work. I joined in September 2026 and am still onboarding, so there is no case study yet.

Onboarding

Internal systems, customer information and unpublished figures are left out for confidentiality.