Case study
Quality Issues Matched Across Two Languages for a Japanese Truck Maker
How SquareShift built a BERT-based NLP model that matches new quality issues to the three closest historical cases in English and Japanese, cutting manual search time.
Book a session →
Client
A leading Japanese manufacturer of commercial trucks, serving domestic and international markets.
Industry
Automotive
Services
Stack
BERT · Machine Translation · PySpark · PyTorch · AWS Lambda
How SquareShift delivered it.
The challenge
The client manufactures commercial trucks for domestic and international markets. Dealers reported new quality issues constantly, and every one of them had to be checked against a large historical database by hand.
Issues were logged separately in English and Japanese, so an engineer working in one language had no easy way to see whether a similar case already existed in the other. Search and comparison were slow, and that delay hurt resolution time and customer satisfaction.
What we delivered
SquareShift translated the Japanese issue reports into English so every case, regardless of source language, could be searched the same way. On top of that shared database, the team built a custom BERT-based NLP model trained to compare a new issue against the full historical record.
For every new quality report, the model now surfaces the three closest matching historical issues automatically, in both languages, through one bilingual search interface.
The outcome
Quality engineers work from one bilingual search experience instead of two separate manual processes. They get the three best-matching historical issues the moment a new report comes in, instead of searching a large database by hand.
Translating and matching through a single model cut the time between a dealer report and a resolved issue, at a manufacturer whose quality-assurance process spans a large international dealer network.
A bilingual quality process usually means two workflows bolted together. Treating an English report and its Japanese twin as one search problem, not two, is what actually saves an engineer's time.
AI & Machine Learning Practice Lead, SquareShift
Related reading: legal judgment summarization to audio and risk signal detection with OCR and machine learning.
