Oct 2022 – Aug 2023 · 10 mos

• Financial Analyst in CIB, JP Morgan Chase

Oct 2022 – Aug 2023 · 10 mos

• Developing regression models to predict revenue streams for various lines of businesses spread across multiple investment tools which are in-turn spread across multiple geographic regions
• Separate models developed for each investment tool for each geographic region which will later on be summed up for macroeconomic view of business
• Revenues either regressed against specific Macro-economic indicator variables or sometimes regressed with itself (case of low correlation with MEV and high possibility of autocorrelation)
• Regression models developed only on stationary series (non-stationary series are made stationary after differencing)
• Once prototype models are developed in excel, the production version is written in R mainly for ease of automation and repeated runs of the models required every month
• These models predict the revenue streams for various investment tool for various geographies for a period of at least 1 year in the future
• Models are validated based on pseudo out of sample errors
• Developed SARIMAX models for forecasting revenues where the exogenous variables are Principal components (PCA) of Macroeconomic variables
• Usage of PCA for exogenous variables helps in removing multicollinearity and also retains the variance in the data

Aug 2020 – Sep 2022 · 2 yrs 1 mo · India

Aug 2020 – Sep 2022 · 2 yrs 1 mo

• Usage of graph and clustering algorithms to find prospect customers based on similarity of patterns observed with respect to existing customers of Bajaj Finance Limited
• Text based standardization of data and spell correction of specific variables. Algorithm developed grounds up and uses unsupervised method of correction and standardization
• Deriving clusters of fraudulent customers through variable encoding and clustering techniques (combinatorial analysis included later for higher efficiency)
• New Algorithm developed for comparing multiple long strings in one shot. The method greatly reduced the computational load and is widely used by analysts across the company
• Graph embedding algorithms (Node2vec) for creating clusters and communities within a graph
• Converting a nominal variable to ordinal variable based on dependent variable distribution (cross tabulation). This helps us to rank entities based on specific attributes
• All the above-mentioned capabilities were completely built on Pyspark using azure databricks as the platform and either EDW (Enterprise data warehouse) or Datalake as the data source and destination
• Total Projects worked:
• o Prospect Offer upgrade project – 10.7 million offer upgrades generated
• o RCU SDK Fraud detection project – Fraud detection based on bank account number patterns
• o RCU ECOM Address clustering pattern – One-shot address comparison
• o Text Standardisation -Basic+Context+Spelling
• o Variable scale conversion - Nominal to Ordinal scale conversion
• o Prospect Graph creation – community detection, graph embedding, MultiGraph (networkX)

May 2019 – Jul 2019 · 2 mos · India

• Summer intern in data analytics. Analysing customer choices in commercial banking.

Jul 2017 – Apr 2018 · 9 mos · Greater Coimbatore Area

• GET in Sales Analytics - Project Phoenix
• Sales and penetration analysis of spare parts of large electric screw air compressors.

Jun 2016 – Aug 2016 · 2 mos · Greater Chennai Area

• Summer Intern in Piping division
• Project : Pipe stress analysis and positioning and replacement of pipe supports.