Sep 2019 – Mar 2022 · 2 yrs 6 mos · Bengaluru Area, India

• Led the computer vision charter working across discovery and catalog problem statements. Was responsible for all aspects of business problem solving through ML starting from Problem identification, model development, KPI identification, model deployment to monitoring and alerting. Joined as the second data scientist in the org and was instrumental in hiring, mentoring and onboarding of data scientists and product analysts.
• Models deployed in production:
• image search
• similar image recommendation
• image-based attribute extraction
• Automated catalogue image quality assessment
• OCR based key value extraction from invoices

Nov 2018 – Sep 2019 · 10 mos · Bengaluru Area, India

• Worked on optimising supply chain using data and machine learning. Created model for inventory balancing between warehouses to reduce stock out events. Implemented model for auto routes generation and allocation to FEs for a given set of orders to be delivered on a particular day.

Jun 2017 – Aug 2018 · 1 yr 2 mos · Bengaluru Area, India

• At Envestnet | Yodlee, I was working on Transaction Data Enrichment engine(TDE) which sits at the core of almost all of Yodlee's revenue generating businesses. The product involves extraction of useful information like merchant name, category, city, state, geolocation etc from raw bank/card transaction description which are very fuzzy in nature. More details at https://www.yodlee.com/products/data-analytics/transaction-data-enrichment/
• Deployed Deep learning technique based models(CNN/RNN for classification and Seq2Seq for tag prediction) into the production where these are run over over millions of transaction daily.
• Resulted in a increase in accuracy as well population rate of merchant and category

Aug 2016 – May 2017 · 9 mos · Bengaluru Area, India

• Developed cross sell and upsel models for AXA Japan entity. The AXA Life Japan(ALJ) - AXA Direct Japan(ADJ) Cross Sell project which was XGBoost model trained on snapshot data resulted in 1.87x lift in top decile. Also the B-policy cross sell project led to resulted in a lift of 1.5x from natural conversion rate.
• Was also actively involved in research group at DIL, Paris and published two studies on class imbalance problem involving analyses of various class balancing sampling techniques and cost based learning.

Aug 2014 – Mar 2016 · 1 yr 7 mos · Hyderabad, India

• Developed a lead prioritization AI engine predicting home loan buyers based on user data submitted online. Implemented Gradient Boosting based models on the features created from the users' data with an AUC score of ~90%
• Created web based dashboards for automation of application defects reporting to project executives.
• Designed and created environment health check dashboard for real time monitoring of all third party interfaces.