The brief

The nonprofit's mission ran on matching talent to opportunity, but the review and matching work was manual and capped by team capacity. Growth meant one of two things: hiring reviewers the budget couldn't support, or building software that could review and match at scale while preserving the human judgment that made placements trustworthy.

As a mission-driven org, it also needed the work to open sustainable revenue — not just cut costs. So the real brief wasn't "add AI." It was: scale the core of the mission, keep it trustworthy, and give the organization something it could build a sustainable model on.

Approach

I started from the business goal — which outcomes had to move — and worked backward to the model and the data, rather than leading with the technology.

• Named the outcomes that mattered first (throughput, match quality, new revenue), then chose the model and data to serve them.
• Built an AI review-and-matching engine on TensorFlow and LLMs with the org's own criteria encoded — not a generic off-the-shelf model.
• Kept humans in the loop where judgment mattered, so the system augmented reviewers instead of replacing their expertise.
• Designed the platform so new services — and the revenue behind them — could be layered on the same matching core.

What I delivered

• An AI review-and-matching engine (TensorFlow + LLMs) integrated into the platform.
• A scoring model encoding the org's real matching criteria, with human-in-the-loop review.
• Two new productized services built on the matching core — each a new revenue stream.
• The data pipeline and infrastructure to run matching at scale.
• A technical roadmap tying each build to a business outcome leadership was tracking.

Outcomes

• ✓Two new revenue streams launched on top of the matching platform.
• ✓A 22% increase in revenue over the engagement.
• ✓Review and matching that scaled past the limits of the manual process — with human judgment preserved.

Stack & methods