AVIJIT BHATTACHARJEE

- Owned the end-to-end ML lifecycle for a loan approval prediction pipeline, including data ingestion and feature engineering from banking and TransUnion (TU) data, model training on AWS SageMaker, real-time inference via AWS Lambda, and continuous production monitoring with data/model drift detection, supporting 10K+ daily credit decisions. - Spearheaded the Agentic Coding project from Proof of Concept (PoC) to Beta rollout, benchmarking leading AI agents like Jules and SWE against junior developer perfor

• Initiated a proof of concept focused on leveraging Retrieval Augmented Generation (RAG) with Large Language Models & NLP. • Developed a Python library that consolidates various hypothesis tests for streamlined development and validation of linear, logistic, time series and AI models, significantly improving the efficiency and accuracy of model validation processes while saving valuable time for other data scientists, model validators and quants. • Single-handedly designed and developed a Python Dash UI, s

• Inclusion & Diversity: Developed an innovative AI engine that could detect bias across HR processes along with its level and direction using available data and statistics, involving both structured and unstructured data analytics. To accomplish the work web scrapping of client’s job posting site was done to analyze gendered word choices using natural language processing (NLP) techniques and secondary research. Then different cohorts were identified for potential bias testing using statistical hypothesis t

• Remaining Useful Life Prediction: Designed an AI/ML based solution to help the organization predict remaining useful life of a water pump to save maintenance cost and penalties due to inappropriate service. To accomplish the task ETL pipeline was built using Airflow, python, pyspark. Then we developed forecasting models using Double Exponential smoothing, ARIMA, FB Prophet, Regression based approaches and VAR. Complete ML lifecycle was maintained using Mlflow. Finally, model was productionized to a server