💳

Semantic Transaction Enrichment

Payment authorization payloads contain cryptic merchant descriptor strings: truncated, typo-ridden, and semantically meaningless to traditional keyword matchers. A fine-tuned LFM2-350M parses these in under 35ms, normalizing merchant names, classifying categories, and extracting locations inline during the authorization hot-path.

Auth-window speed — Sub-35ms inference fits inside the 50ms auth window. No Plaid cloud round-trip needed
Semantic understanding — Understands "SQ *JOES CFF SHP NY" means Joe's Coffee Shop, New York. Regex only sees random characters
No cloud dependency — On-prem parsing. No per-call Plaid API cost, no cloud latency in the auth hot-path

The Problem

'SQ *JOES CFF SHP NY': regex can't parse this. Plaid Enrich covers 90%+ but is a cloud API outside the 50ms auth window.

How LFM Compares

Regex can't parse messy merchant descriptors like 'SQ *JOES CFF SHP NY.' Cloud enrichment APIs add latency outside the auth window. LFM parses semantically at <35ms, on-prem.

What LFM Unlocks

Semantic merchant parsing at <35ms, inside the auth window. No cloud round-trip. New POS formats via LEAP in minutes.

Semantic Transaction Enrichment

Parse chaotic merchant descriptors into structured data within the 50ms auth window

Merchant Descriptor Parsing

Card networks limit merchant descriptors to 22 characters — truncated, abbreviated, and cryptic. Traditional keyword matching loses the semantic meaning. Select a descriptor below to see semantic parsing in action.

Select a merchant descriptor:

This demo is fine-tuned on sample data. Results improve with your data.

Build Your Model on Workbench →Talk to Our Team