B2B Integration Payload Classification using Fine-Tuned BERT

An NLP classification pipeline leveraging Hugging Face Transformers (bert-base-uncased) to automatically parse and classify unformatted middleware exception logs (such as EDI 850/810 parsing errors, network timeouts, and authentication failures).

Overview

• Objective: Automate incident triage for enterprise B2B middleware transaction errors.
• Model: Fine-tuned BERT (bert-base-uncased) for sequence classification.
• Tech Stack: Python, PyTorch, Hugging Face Transformers, Pandas, Scikit-learn.

Project Structure

• edi_exception_logs.csv — Synthetic dataset containing unformatted middleware execution and parsing error logs.
• bert_log_classifier.py — Python script for text tokenization, model initialization, and log classification.

Key Outcomes

• Successfully tokenizes and extracts intent from raw EDI transaction error messages.
• Categorizes exception logs into distinct operational buckets (Data_Format_Error, Network_Timeout, Auth_Failure) with high classification accuracy.