Problem
New officers lose real time hunting acronyms, and the reference PDFs are scattered across the intranet and barely searchable.
How it works
Full-text search with fuzzy matching, Jaro-Winkler for typos and Metaphone for phonetic near-misses. I worked through Levenshtein, Soundex, and Aho-Corasick before settling on the pair that handled real searches best.
I tested Valkey for caching the expensive queries, but SQLite turned out faster at this scale, so it stayed. I also looked at MinHash and Bloom filters for scaling, though the dataset doesn't need it yet.
Admin tools find and merge the duplicate entries that creep in over time.
A look inside
