Moody's Corporation
About This Project
Moody's publishes millions of articles weekly across global financial ratings, on a system built in legacy C and C++. I led an 8-week applied AI pilot with the engineering team: discovery research, hands-on workshops, and a human-in-the-loop assistant the engineers built themselves to modernize toward React without taking the product offline.
The Challenge
The system behind Moody's ratings publishing carried years of technical debt in legacy C and C++ code, with no scalable framework for managing it. Modernizing toward a React-based infrastructure was the goal, but the product never stops publishing, so nothing could be shut down to rebuild it.
The engineering team was new to applying AI in their work. My job wasn't to implement the solution. It was to get the engineers to one, using AI, in a way they would trust and keep using after I left.
The Approach
I started with discovery interviews with engineers, the product owner, and the project manager, mapping their workflows, learning their technical language, and finding where the real friction lived. I used AI to compress weeks of ramp-up into days so I showed up to every conversation with enough context to be useful.
A non-engineer in an engineering room is a setup for a short conversation. I opened the first workshop by being upfront about who I was and who I wasn't, and the homework earned the rest.
From there I ran three workshops built for this team, not generic AI literacy, but applied directly to their craft with the tools they'd actually use: Microsoft Copilot, Azure AI, Cursor, Claude Code, Windsurf, and Stagewise. Together we built a hybrid human-AI decision framework: what the tools handle, where a human stays in the loop, and what never gets handed off.
The Outcome
The workshops turned into a build. I helped design and spec a semi-agentic assistant with a human in the loop, and the engineers built and shipped it within the 8 weeks. It gave the team a systematic way to manage modernization work instead of reacting to it:
Jira integration tracked work in real time as it moved
Knowledge base stayed current as decisions evolved
Bug monitoring surfaced issues that needed human review
Living system updated continuously after every sprint
Sprint velocity on the modernization effort rose 27%. The team left with a working tool, a decision framework, and a way of working with AI they built themselves.
Moody's Corporation
About This Project
Moody's publishes millions of articles weekly across global financial ratings, on a system built in legacy C and C++. I led an 8-week applied AI pilot with the engineering team: discovery research, hands-on workshops, and a human-in-the-loop assistant the engineers built themselves to modernize toward React without taking the product offline.
The Challenge
The system behind Moody's ratings publishing carried years of technical debt in legacy C and C++ code, with no scalable framework for managing it. Modernizing toward a React-based infrastructure was the goal, but the product never stops publishing, so nothing could be shut down to rebuild it.
The engineering team was new to applying AI in their work. My job wasn't to implement the solution. It was to get the engineers to one, using AI, in a way they would trust and keep using after I left.
The Approach
I started with discovery interviews with engineers, the product owner, and the project manager, mapping their workflows, learning their technical language, and finding where the real friction lived. I used AI to compress weeks of ramp-up into days so I showed up to every conversation with enough context to be useful.
A non-engineer in an engineering room is a setup for a short conversation. I opened the first workshop by being upfront about who I was and who I wasn't, and the homework earned the rest.
From there I ran three workshops built for this team, not generic AI literacy, but applied directly to their craft with the tools they'd actually use: Microsoft Copilot, Azure AI, Cursor, Claude Code, Windsurf, and Stagewise. Together we built a hybrid human-AI decision framework: what the tools handle, where a human stays in the loop, and what never gets handed off.
The Outcome
The workshops turned into a build. I helped design and spec a semi-agentic assistant with a human in the loop, and the engineers built and shipped it within the 8 weeks. It gave the team a systematic way to manage modernization work instead of reacting to it:
Jira integration tracked work in real time as it moved
Knowledge base stayed current as decisions evolved
Bug monitoring surfaced issues that needed human review
Living system updated continuously after every sprint
Sprint velocity on the modernization effort rose 27%. The team left with a working tool, a decision framework, and a way of working with AI they built themselves.
Moody's Corporation
About This Project
Moody's publishes millions of articles weekly across global financial ratings, on a system built in legacy C and C++. I led an 8-week applied AI pilot with the engineering team: discovery research, hands-on workshops, and a human-in-the-loop assistant the engineers built themselves to modernize toward React without taking the product offline.
The Challenge
The system behind Moody's ratings publishing carried years of technical debt in legacy C and C++ code, with no scalable framework for managing it. Modernizing toward a React-based infrastructure was the goal, but the product never stops publishing, so nothing could be shut down to rebuild it.
The engineering team was new to applying AI in their work. My job wasn't to implement the solution. It was to get the engineers to one, using AI, in a way they would trust and keep using after I left.
The Approach
I started with discovery interviews with engineers, the product owner, and the project manager, mapping their workflows, learning their technical language, and finding where the real friction lived. I used AI to compress weeks of ramp-up into days so I showed up to every conversation with enough context to be useful.
A non-engineer in an engineering room is a setup for a short conversation. I opened the first workshop by being upfront about who I was and who I wasn't, and the homework earned the rest.
From there I ran three workshops built for this team, not generic AI literacy, but applied directly to their craft with the tools they'd actually use: Microsoft Copilot, Azure AI, Cursor, Claude Code, Windsurf, and Stagewise. Together we built a hybrid human-AI decision framework: what the tools handle, where a human stays in the loop, and what never gets handed off.
The Outcome
The workshops turned into a build. I helped design and spec a semi-agentic assistant with a human in the loop, and the engineers built and shipped it within the 8 weeks. It gave the team a systematic way to manage modernization work instead of reacting to it:
Jira integration tracked work in real time as it moved
Knowledge base stayed current as decisions evolved
Bug monitoring surfaced issues that needed human review
Living system updated continuously after every sprint
Sprint velocity on the modernization effort rose 27%. The team left with a working tool, a decision framework, and a way of working with AI they built themselves.
Moody's Corporation
About This Project
Moody's publishes millions of articles weekly across global financial ratings, on a system built in legacy C and C++. I led an 8-week applied AI pilot with the engineering team: discovery research, hands-on workshops, and a human-in-the-loop assistant the engineers built themselves to modernize toward React without taking the product offline.
The Challenge
The system behind Moody's ratings publishing carried years of technical debt in legacy C and C++ code, with no scalable framework for managing it. Modernizing toward a React-based infrastructure was the goal, but the product never stops publishing, so nothing could be shut down to rebuild it.
The engineering team was new to applying AI in their work. My job wasn't to implement the solution. It was to get the engineers to one, using AI, in a way they would trust and keep using after I left.
The Approach
I started with discovery interviews with engineers, the product owner, and the project manager, mapping their workflows, learning their technical language, and finding where the real friction lived. I used AI to compress weeks of ramp-up into days so I showed up to every conversation with enough context to be useful.
A non-engineer in an engineering room is a setup for a short conversation. I opened the first workshop by being upfront about who I was and who I wasn't, and the homework earned the rest.
From there I ran three workshops built for this team, not generic AI literacy, but applied directly to their craft with the tools they'd actually use: Microsoft Copilot, Azure AI, Cursor, Claude Code, Windsurf, and Stagewise. Together we built a hybrid human-AI decision framework: what the tools handle, where a human stays in the loop, and what never gets handed off.
The Outcome
The workshops turned into a build. I helped design and spec a semi-agentic assistant with a human in the loop, and the engineers built and shipped it within the 8 weeks. It gave the team a systematic way to manage modernization work instead of reacting to it:
Jira integration tracked work in real time as it moved
Knowledge base stayed current as decisions evolved
Bug monitoring surfaced issues that needed human review
Living system updated continuously after every sprint
Sprint velocity on the modernization effort rose 27%. The team left with a working tool, a decision framework, and a way of working with AI they built themselves.