Plain English to 40,000+ SKUs.
Customer descriptions rarely match an internal catalogue cleanly. I built a browser-based JavaScript application that combines rule-based parsing, normalization, fuzzy and phonetic matching, and confidence scoring to suggest the right product code.
In informal internal testing, it correctly resolves approximately 85% of descriptions and returns no result when confidence is insufficient. Incorrect matches have been rare, and an organization-wide second release has been approved.