Generate curated catalogue from noisy product data of Micro, Small & Medium Enterprises. Curated catalogue would help MSMEs to standardize product data, thereby, making accounting and point of sale more efficient.• Designed and implemented a scalable catalogue pipeline which takes product data as input and generates clusters containing similar representation of same product as output. The pipeline was robust to common sources of noise present in product data like misspellings, abbreviations and missing attributes.• The pipeline involved nearest neighbour search using locality sensitive hashing on product embeddings. Results were further filtered based on lexical features and attribute features like brand, category and unit of measurement extracted from product descriptions.• Conceptualized and implemented a token correction model which auto corrected misspelled and abbreviated words present in product titles. Correct word suggestions were obtained using Word2Vec and FastText models with affine gap distance threshold. Built a word level language model using LSTM networks to choose correct token based on context of product descriptions.

• Intermittent Demand Forecasting Model - Forecasted weekly demand patterns of 1000 Stock Keeping Units using intermittent demand forecasting methods. Built predictive model which used best performing algorithm out of Multiple Aggregation Prediction Algorithm (MAPA) and Croston method and its variants; Performance was tested on out of time, walk forward validation set data• Smooth Demand Forecasting Model- Built XGBoost model with lagged auto regressive terms for product categories demand; Forecasts were split into customer segment demand quantities by forecasting based on historical ratios; Ratios were forecasted using ARIMA• Descriptive Model - Built GBM model to study effects of external variables such as weather, vehicle registrations and google trends on demand. Performed feature engineering to derive multiple rolling window features; Used grid search to tune hyper parameters. Created partial dependence plots to understand relationship between predictors and demand at product category levels

• Transaction Monitoring Models - Built transaction monitoring model for entities external to bank, this model was to be used for 55 countries. Performed cluster analysis to segment customers and accounts based on transaction activity. Received Delivering the Promise (Q2'15 & Q2'16), Team Star(Q4'15) & Leading light (Q2'16) awards.• Validation of thresholds - Led a team of 4 analysts to provide enhanced solution for validation of monitoring thresholds. Achieved an increase in efficiency of process by 12% and effectiveness by 3% over existing solution.