Akshat Kumar Email & Phone Number

Bentonville, Arkansas, Us

• Led an end-to-end project to develop an automated data pipeline for predicting employee attrition, including creating associate profiles, training and tuning tree-based models, and implementing a feedback loop to track interventions. Delivered actionable insights via Power BI dashboard, empowering senior leadership to make data-driven decisions, which led to a 12% reduction in attrition.• Estimated the causal impact of wage hikes on employee turnover using fixed effects and causal ML models, incorporating segmentation and policy analysis to provide HR and People Partners with tailored insights for optimizing investment strategies.• Developed a generative AI solution for automated text summarization, leveraging GPT-based models to summarize annual engagement survey feedback from 100,000 associates across 600 stores, identifying key themes and sentiments. Created an interactive dashboard to address critical issues and enhance employee engagement and organizational culture.

Berkeley Heights, Nj, Us

Led onshore forecasting team in sales predictions of major anti depressant drugs and analyze dynamics that impact brand performance• Time series based supervised learning model developed using physician’s and patient’s characteristics, personal and non personal promotions, COVID impact, competitor’s drug sales, market growth indicators, auto-regressive sales lags and seasonality• Devised and implemented a top to bottom split method to yield the national forecast at multiple sub-national levels• Final optimized model framework yielded 6% less error on average in terms of overall volume and month on month sales predictions in comparison to client’s own internal forecasting technique

Franklin, Ma, Us

Developed an end to end dialog state classification component for the company’s Interactive Voice Response Platform• Classified 5000+ dialog states spanning over 50+ voice applications in the Impala data lake at Interactions and subcategorized these types for deeper insights• Statistical rules based random forest and textual features based bidirectional LSTM DNN models built for dialog state classification• Developed automated scripts that pulled data at specified intervals from Impala datalake, loaded them into the trained classifier and populated the predicted result back to the database• Classification data used for generating reports on automation and analyzing performance across applications by dialog state type

Bangalore, Karnataka, In

Designed and developed scalable statistical models that solved complex client use cases• Formulated and implemented Unsupervised Binning algorithm for traveler type classification for an online travel company. Model processed close to 0.8 million records per second on IBM PPC S822L machines• Built De-Duplication model on top of Apache Spark that reduced cost of storage infrastructure and maintained uniqueness with an accuracy of 92.59 % in the client database.• Proposed and implemented Most Valuable Player Rank Discovery algorithm that churned out the best strategy for cricket teams in terms of player selection.• Performed ARIMA based time series analysis for ATM dispense cash forecasting. The model generated a low RMSE of 16.41 indicating it accurately predicted the currency forecast demand• Built Agglomerative Hierarchical Clustering solution that created cluster hierarchy of Hindi language TV viewers for an Indian TV viewership measurement organization

Delhi, New Delhi, In

Worked on image classification of a remote sensing image of India through discretization, reduct identification and subsequent rule generation• Discretization algorithms applied on the 7 band image dataset to observe the effects of discretization on the accuracy of classifier• Boolean Reasoning model classified the area most distinctively with 94.37 % accuracy• Research got published as the first author at the 2nd IEEE International Conference on Computer Communications and Systems (ICCCS-2014)

Master Of Science - Ms, Data Science

Bachelor'S Degree, Computer And Communication Engineering