Case study
Reporting agent
Microsoft internship project where I built an agent that analyzed product metrics, produced a weekly report, and notified team members of significant metric changes. Cleansed for portfolio use.
The problem
During my Microsoft internship, I worked on a reporting process that was valuable, but time-consuming. Each cycle required people to collect product metrics, review changes, identify meaningful movement, write narrative updates, and notify the right owners when something needed attention.
The goal was to move the process from manual assembly toward a repeatable agent workflow: less time spent gathering information, more time spent understanding what changed and what action should happen next.
How I approached it
Requirements first
I started by understanding what the report needed to accomplish, which sections mattered most, what data needed to be trusted, and where human review still belonged in the process.
Stakeholder input
I talked with team members to understand metric ownership, review expectations, and where the current process created friction or repeated follow-up.
Agent instructions
I learned how skill and instruction files shaped the agent's behavior, then used them to define the report structure, writing rules, data inputs, and execution steps.
Iterative build
I built toward a working foundation first, then layered in improvements like owner routing, change callouts, repeatable configuration, and safer review paths.
How the agent works
Load the reporting period, team configuration, owner mapping, and current and historical metric inputs.
Compare metric movement over time, identify notable changes, and connect each area to an owner or review path.
Generate the report structure, metric tables, plain-language summaries, and callouts for areas that need attention.
Send the output for review and trigger notifications when significant changes need follow-up from the right people.
Tools and product decisions
The key product decision was keeping the system configurable, so report structure, owners, data sources, and workflow steps could change without rebuilding the entire agent.
Impact to users
Saved 4–5 hours a week on reporting.
Reduced manual effort
Automated repetitive report assembly and metric review steps.
Improved consistency
Standardized how metrics, owners, follow-ups, and changes were captured.
Stronger signal
Highlighted meaningful metric movement instead of only producing status updates.
Reusable foundation
Created a configurable foundation for future reports and teams.
What I learned
This project helped me connect product management and AI implementation in a practical way. I had to translate messy reporting needs into requirements, understand enough of the technical system to shape the agent, and make product decisions about trust, review, usability, and impact.