Back to case studies
Case Study · 03

Product Classification System

Transformer-based classifier turning unstructured product descriptions into structured categories at enterprise scale.

Architecture

Descriptions in, categories out.

A clean inference path from raw product descriptions to structured category outputs, deployed through ONNX Runtime and a .NET API.

classification.pipeline · inference path
2M+ / month
01
Product descriptions
unstructured text
02
Preprocessing
clean · normalise · tokenise
03
Transformer model
fine-tuned encoder
04
ONNX Runtime
optimised inference
05
.NET API
enterprise integration
06
Category output
600+ classes
transformer.predict(description) → category · confidence 0.97 · latency ~12ms · runtime onnx
Impact

Production numbers.

2M+ / mo
Predictions served
600+
Product categories
95%+ F1
Top 85% of products
How it works

From data to enterprise inference.

Anonymised summary of the ML delivery flow and deployment stack.

01

Problem

A global maritime client needed to turn unstructured product descriptions into a strict taxonomy of 600+ categories at enterprise scale.

02

ML pipeline

Owned end-to-end delivery: data gathering, labelling analysis, experimentation, model evaluation, and error analysis on long-tail categories.

03

Model deployment

Fine-tuned transformer exported to ONNX Runtime for fast, portable inference - served behind a .NET API embedded in the client's stack.

04

Enterprise scale

Serving 2M+ predictions per month with predictable latency, integrated into existing enterprise data flows and monitoring.

05

Performance

95%+ F1 score on the top 85% of products, with error analysis feeding continuous data and labelling improvements on the long tail.