- Created an Anomaly Detection framework to identify user transactions with incorrect prediction of associated merchant from enriched transactional data, using pySpark to query and filter over a billion records.- Developed a framework to predict merchant for anomalous transactions using Google SERP API and trained a logistic regression classifier to determine the accuracy of predicted merchant.- Created a model to give an accuracy score to the transaction category from enriched transactional data. Final model was XGBoost Classifier with around 90% accuracy which was tested against other ML models including SVM, Random Forest, ANN etc.- Worked on creating a framework to generate possible identifier keywords present in transaction descriptions for a given list of high volume merchants, which helped in improving both precision and recall for merchant prediction.- Worked on developing a module to predict Store ID for a given transaction using 3rd party database, which helped in improving the enrichment quality of approx. 35% transactional data.- Worked on developing a module to predict associated merchant for transactions based on the presence of phone number or address in the txn. description using 3rd party database, which helped in improving recall of merchant prediction during enrichment of transactional data by about 15%.

- Created a graph database using Neo4j from 850k transactional data of around 10k people, which can be queried using Py2neo library in python.- Created a dashboard built on top of graph database using Dash module in python. It helped in visualization of spend patterns of users and identifying user cohorts/segments.