Email Conversation Bot Development (BERT Fine-Tuning): Fine-tuned a pre-trained BERT small model to automate email responses, increasing productivity and reducing manual intervention in routine email handling. Developed a custom dataset and adapted the BERT model to classify emails into categories .Achieved a classification accuracy of 91.5%, demonstrating strong model performance and robustness. Implemented a rule-based response Anicalls generation system to automatically reply to emails based on predicted categories. Named Entity Recognition (NER) Model (BERT Fine-Tuning): Built a highly accurate NER model using fine-tuned BERT to extract key entities from unstructured text, such as leader names, hotel names, and cities. Collected and labeled diverse datasets to train the model, enabling entity extraction from hotel booking pdf. Evaluated model performance with precision, recall, and F1-score metrics, ensuring robustness across various domains. Successfully optimized the model with hyperparameter tuning and regularization to prevent overfitting and improve accuracy. Conversational bot Using GPT-4O-Mini: The bot interacts with hotel agents and cross-checks booking details, reducing manual labor and improving operational workflows for booking re-verification teams. The solution provides a more reliable and faster response system for booking reverification team, contributing to higher customer satisfaction and better management of re-verification operations. Model Optimization and Hyperparameter Tuning: Leveraged advanced optimization techniques such as learning rate adjustments, batch size modifications, and sequence length fine-tuning to enhance model performance. Applied crossvalidation to ensure generalization across multiple data sets and domains. Ensured robust evaluation through the implementation of standard NLP metrics and iterative model refinements.

• Prediction Explanation Model (NLP, LIME, SHAP):Developed a prediction explanation model using LIME and SHAP, improving the interpretability of machine learning models for enhanced decision- Anicalls making and transparency. Applied LIME (Local Interpretable Modelagnostic Explanations) and SHAP (Shapley Additive explanations) to explain complex model predictions in a human-understandable manner. Enabled stakeholders to trust and effectively use the model by providing insights into how input features influence predictions. • Data Visualization and Reporting: Created advanced data visualizations to transform complex datasets into easy-to-understand graphical representations. Designed charts, plots, and interactive dashboards to visualize trends, patterns, and insights for both technical and non- technical stakeholders. Improved decisionmaking and reporting by presenting data in a compelling and clear format. • Clustering and Segmentation (Unsupervised Learning): Led a segmentation task, applying various unsupervised machine learning algorithms to form distinct clusters from raw data. Applied techniques such as K-means, DBSCAN, and hierarchical clustering to identify hidden patterns and group similar data points. Enhanced the model's ability to discover meaningful data structures, resulting in improved segmentation accuracy