Key Responsibilities: Designed a SMS classifier and event extraction service using text embeddings, hierarchial XGBoost classification and NER pipeline, achievening 95% accuracy in classifying SMSs into 20+ categories and extracting relevant financial events from user data. Extracted user's credit lines and related financial events with 85% Recall and 90% Precision across defaults and creation of new credit lines for users with available data, ahead of Bureau report. lending-related decisions.

Technologies: Python, SQL, Spark, AWS (EC2, Lambda, S3, SQS), Sentence Transformers, Sentence Embeddings and Machine Learning Libraries.

Impact: Enabled the prediction of potential borrower risk and accelerated critical lending-related decisions.

Key Responsibilities: Developed a shipping cost allocation tracking tool for the TFS team that collects partial allocations across services and displays in a common interface. This reduced the manual effort in cost component querying by up to 80% and optimised the time required for on-call engineers to detect cost allocation anomalies and trace the root cause of errors by up to 90%. This tool streamlined the process by automating cost tracking across various upstream services and the final allocated cost.

Tech Stack: Java, Python, SQL, Spark, AWS (EC2, S3, Lambda, DynamDB, SQS, SNS), REST API and Frontend Frameworks

Impact: Streamlined the process of error debugging across owned services and reduced the time to identify and resolve issues.

Project Highlights: Architectured and implemented an end-to-end module that streamlined goal management for automated monitoring of supply chain data, ensuring real-time responsiveness to critical metrics.

Tech Stack: Java, Python, REST API, AngularJS, NodeJS.

Impact: Made goal management easier and faster for the team.

Program Focus: Received intensive mentorship in software engineering best practices from Microsoft engineers. Developed a recommendation engine that implements Content-Based Filtering and Collaborative Filtering using using KNN Clustering and Similarity Searching to learn the fine interactions between users and movies. User Based Model achieved RMSE of 1.5 and demonstrated 50% more efficiency and better recommendations compared to Item Based Model which had RMSE of 2.5.

Skills Developed: Python, Streamlit, Similarity Search, KNN Clustering, Machine Learning Libraries.

Outcome: Successfully completed program with a return offer.