AI Merchant Name Translation
Cyrillic, Greek, Arabic merchant strings in your bank statements? The on-device LLM transliterates and translates them into searchable English keywords.
Why travel statements break expense search
Travel statements come back full of "АТБ" or "Καρρέ" — useless for analytics search. Budgie's translation pipeline runs the merchant string through the on-device LLM, gets a normalized name plus search keywords, and stores both.
Original strings stay intact for receipt-matching; the translated form is what your search queries hit. Multi-script — Cyrillic, Greek, Arabic, CJK — all supported.
What you get
Cyrillic, Greek, Arabic, CJK, and more — any non-Latin script the on-device LLM understands
Original merchant string preserved for receipt-matching and audit
Translated form indexed by embeddings — semantic search hits whatever you remember
On-device — no merchant string ever leaves your phone
Manual override per merchant if the LLM gets it wrong
How it works
Each new transaction's title flows through a translation queue. The LLM proposes (translatedName, keywords); both are saved alongside the original. Embeddings then index the translated form for semantic search.
Frequently Asked Questions
Which scripts are supported?
Are the original merchant strings kept?
What if the LLM mistranslates?
Does this run on every transaction?
Related Features
Read More on the Blog
Ready to take Budgie for a spin?
Join the waitlist — be first to try the offline-first expense tracker.