I built a custom tool that uses natural language processing and a proprietary algorithm—built from the ground up with cosine similarity and weighted scoring—to suggest redirects for broken URLs based on the root keywords in each url path. If the model doesn’t find a match above the defined threshold, it applies a series of cascading rules to generate fallback recommendations. The client needed a highly tailored solution, and this approach proved more precise and aligned with their site structure than my standard redirect generator.
Given the site’s size and complexity, I added a dynamic bar chart to highlight problematic directories and support intuitive navigation. Users can drill down through the chart, click through nodes in the visualization, or use breadcrumbs to explore and return to higher levels.
To make the redirect logic more transparent, I exposed the similarity score in the results table, which is also searchable and updates in real time as users move through the architecture.
Finally, I hosted it on my site so the team can interact with the data and added an export button (not shown) so the team can export their current view to csv.
The broader goal is to intelligently eliminate broken pages across a site. But what made this project especially rewarding was designing and implementing the cascading logic that makes the whole system tick.
Once they signed off on the redirect strategy, I generated a redirect map specific to their server that they could lob it over the fence to their engineers to rid their site of broken pages.
👋 ⛓️💥⛓️💥⛓️💥