SUBHASH BUNKAR

Unraveling Customer Chatbot Utilizing T5: Extracted and cleaned Unravel documentation text from official website using BeautifulSoup to enhance the usability. Developed a chatbot to assist the customers in identifying configuration features & basic information about unravel’s recommendations utilizing Langchain and Llama2 LLM. Unravel Real-time Query Calculator: Extracted features from SQL Queries using SQLGlot to construct a regression-based model for SQL query cost prediction prior to execution. Trained ML models such as linear regression, Random Forest and XGBoost regression and achieved an Adjusted R^2 of 0.85. Trained an XGBoost classifier model with 84% recall & 78% precision to categorize queries into 3 categories (Low, Medium, and High). Healthcheck: Performed EDA of query, warehouse insights data & generated backend health check report json to assist customers in decision-making regarding the Unravel product. Others: Developed and deployed 2 dashboards using Plotly and Dash for cost anomaly of the EMR user’s data. Developed a ML model utilizing LR, Decision Tree, SVM Random Forest, XGBoost, Adaboost and Neural Networks to categorize the high and low-cost Spark applications. Optimized BigQuery cloud cost by generating custom quotas suggestions and moving queries to a different pricing plan.

Developing utilities (Python, OOPs) and handling the IEP pipeline which is used to Priorities the Indication for a given target (Protein). Developed market analysis component to Priorities Indication based upon current marketed and clinical trial status. Developed a CNN model (Recall: 87%, Precision: 78%) for predicting gene-disease association using Protein sequence & Disease Symptoms embedding.

Scrapped reviews of 10 different 'KAPIVA' products from Amazon using Web Scrapper and analyzed for model creation. Developed a model with hyperparameter tuning using Grid search CV to categorize reviews into positive and negative classes using LR, RNN, LSTM, and BERT. Identified 40-50 relevant keywords of the KAPIVA products to optimize AMAZON listing.

Selected important features and combined them with affinity data to predict offer types for the lead to maximize closure rate. Developed a ML model and achieved 87% recall and precision 83%, used KNN, DT, Random Forest, Adaboost and XGboost to predict offer for daily leads. Predicted probability of lead's conversion using logistic regression & took monthly lead closure rate from 1% to 15% in 2 months.