The brief
MatchYourSaaS compares software tools for buyers who are tired of review-site noise. The v1 needed a real comparison engine: a normalized data model for products and features, a scoring layer, and an editorial surface a non-engineer could maintain.
That is not a sprint. It is a continuous workload across many connected features. We embedded one senior engineer for eight weeks.
The constraint
Comparison data goes stale fast and wrong data is worse than missing data. The engine had to make it easy to keep products current and hard to publish a comparison built on a gap.
The approach
We modeled products, features, and feature values as first-class records, then built a scoring layer on top that is fully explainable. Every score in a comparison can be traced to the underlying feature values that produced it.
Claude is used in one narrow place: drafting the prose summary of a finished comparison from the structured scores. It never invents a score and never touches the data model. A human approves every published summary.
The build
Weeks 1 and 2: the data model and the admin surface. Weeks 3 to 5: the scoring layer and the comparison pages. Weeks 6 and 7: the Claude summary drafting with a human approval step. Week 8: hardening, performance, and handoff.
The outcome
Comparison engine v1 went live at the end of week 8. MatchYourSaaS can now publish a new comparison without an engineer, and every score on the page is explainable down to the feature value that produced it.
“Embedded meant I got an engine, not a feature. Weekly cadence, monthly written review, every score explainable. It felt like having a senior engineer on staff without the 12-month contract.”