NLP e2e Industry Level Project Part - 5 | Model Evaluation | Inference Pipeline | Complaint Routing

In this final part of the industry-level NLP project, we complete the full pipeline by evaluating the trained model, building the inference pipeline, and implementing the automated complaint forwarding system. This video concludes the end-to-end NLP project where customer complaints are classified and automatically routed to the appropriate department. Here is the GitHub repo link: https://github.com/switch2ai You can download all the code, scripts, and documents from the above GitHub repository. In previous parts of the project, we built the entire machine learning pipeline including data gathering, exploratory data analysis, text preprocessing, tokenization, embedding representation, model training using LSTM, handling class imbalance using class weights, and building the inference pipeline. In this final stage, we focus on model evaluation and real-world application logic. First, we load the trained LSTM model and the tokenizer that were saved during the training phase. The tokenizer ensures that incoming complaint text is converted into the same token IDs used during training. Next, we evaluate the model using predictions on the test dataset. The model outputs probability scores for each department category. These probabilities are converted into predicted labels using argmax. We then compare predicted labels with actual labels to evaluate model performance. A confusion matrix is generated to visualize how well the model performs across different department categories such as Card, Credit Report, Loan, Others, and Services. We also generate a classification report that provides precision, recall, F1-score, and support for each class. This helps understand how well the model performs for minority classes and ensures that the system generalizes properly. After evaluation, we build the inference pipeline used for real-time predictions. Incoming complaints are first cleaned using the same preprocessing function used during training. The complaint text is co