Shashikant Singh

Worked on next Gen Artificial Intelligence technologies, including but not limited to Transformers, GPTs, OpenAI, LangChain, HuggingFace, LLMs, Fine Tuning of LLMs. Developed and deployed a Customer Sentiment prediction machine learning model on Azure Databricks utilizing MLOps including model experiments, model logging, model registry, achieved 90% accuracy which resulted in enhanced product ratings by 10 % utilized SiBERT model from HuggingFace. Worked on multiple Natural Language Processing (NLP) models both using traditional machine learning and latest LLMs using PyTorch and Python. Fine-tuned LLM models. Actively participated in multiple competitions on Kaggle on machine learning, deep learning, image classification, NLP etc. Collaborated closely with teams, business stakeholders and engineers to ensure smooth integration of models on production environments. Well versed with Large Language Models (LLMs) and their fine tuning methods. Led a team of 2 Data Scientists and 1 Data Engineer in developing Multi-Touch Attribution Model that led to 21% reduction in cost per lead. Achieved €2.8 million in operational cost savings for a Fortune 100 company by developing and optimizing multi-variate hierarchical time series forecasting models utilizing feature engineering and hyperparameter tuning with 14% reduction in forecast error. Led a team of 7 for development of comprehensive end-to-end stability study and CPK analysis product for a Fortune 500 pharmaceutical client based on regression model. Identified trends in Control charts and Stability analysis for Mebeverine and saved client $10 million by reducing scrap batches. Performed K-means clustering on time series data , reduced forecasting time by 40%. Acted as an SME and thought leader, including the creation of project timelines, interviewing data scientists and bringing the best talent to MathCo. Mentored Data Science freshers on Demand Forecasting Capstone project.

Increased client retention by 33% by building a classification machine learning model to predict retention with an accuracy of 75%. Developed a deep learning image matching model using CNN, to match products across different competitors for a client with an accuracy of 72%. Collaborated with stakeholders to understand problem statement, refined the scope of analysis, and used the results to drive informed decisions. Automated data cleaning processes with Python scripts, improving data quality by 25% and reducing preparation time for analytical tasks. Refactored legacy code to improve reliability, scalability and maintainability. Mentored Data Science students of Cornell University on their Capstone Project in QL2 Software.

Reduced credit card default rate by 12% by building propensity classification model to predict default with an accuracy of 82%. Developed a regression machine learning model to predict credit card limit for different individuals based on different parameters. Analyzed large data sets and assisted in the development of custom models/algorithms to uncover trends, patterns and insights. Performed time series analyses for different products of the bank. Analyzed business data to produce reports on PowerBI and polished presentation, highlighted findings and recommended changes. Developed clean and organized machine learning codes using standard data science methodologies. Involved in parsing and aggregating messy, incomplete, and unstructured data sources to produce datasets that can be used in analytics/predictive modeling. Developed an ETL pipeline on AWS using python scripts to automate tasks.

Developed a customer churn classification model for a client which predicted churn with an accuracy of 67%. Worked with clients to understand their business objectives and leveraged profound understanding of mathematical and statistical principles to deliver data driven solutions. Implemented business intelligence solutions with Tableau to improve data-drilling capabilities for non-technical users. Reduced report generation time by 35% by automating SQL scripts.