• Forecasted Wind Power at farm level for renewable energy major for next 1.5 hrs containing 119 turbines using Deep learning model Conv-LSTM using TensorFlow reducing penalty by 8% . Solved the complex wake effect between turbines capturing spatial-temporal relationship between time series. Built an end to end DeepAR framework to forecast the irradiance of 69 solar farms for the next 48 hours and predict for related but unknown timeseries with better accuracy• Using deepface… Show more • Forecasted Wind Power at farm level for renewable energy major for next 1.5 hrs containing 119 turbines using Deep learning model Conv-LSTM using TensorFlow reducing penalty by 8% . Solved the complex wake effect between turbines capturing spatial-temporal relationship between time series. Built an end to end DeepAR framework to forecast the irradiance of 69 solar farms for the next 48 hours and predict for related but unknown timeseries with better accuracy• Using deepface identified dominant emotions of facial images for Philippines based banking client and evaluated significance of different facial embeddings algorithms including VGG-FACE,FACENET etc. for a sample dataset. Created important facial features using dimensionality reduction (PCA) and visualized in t-SNE. Identified regions of model activation in facial images and model bias using guided GRAD-CAM visualizations• Built a credit risk classification model for Quick and Flex loan application for a banking client with 0.4 Gini Index through curated features and feature transformations using ensemble modeling and stacking models with Xgboost, Logistic regression and Random Forest. Identified customer profiles and preferred time slots to call through statistical tests and clustering improving collection strategies of call centers.• Developed a GPT-35 powered LLM based autonomous agent to help CMO of CPG company to optimize the market spends and answer business questions with 75% accuracy for an AI-challenge. Developed Langchain based custom SQL Agent and Python Agent to answer structured data and generate codes. Implemented RAG for answering questions based on unstructured data. Integrated a custom prompt to redirect questions to the required tools and custom output parser to monitor the agent output. Show less

• Spearheaded the time series forecasting for a restaurant chain in USA using regression models .Scaled the modeling pipeline for 2000 cafes across 4 channels in GCP using PySpark framework resulting with overall wmape of 15%.Engineered features based on history and cafes operating minutes along with oversampling for capturing extreme event predictions including Halloween, Christmas etc. Developed a explainable framework using feature coefficients for forecasting prediction• Developed… Show more • Spearheaded the time series forecasting for a restaurant chain in USA using regression models .Scaled the modeling pipeline for 2000 cafes across 4 channels in GCP using PySpark framework resulting with overall wmape of 15%.Engineered features based on history and cafes operating minutes along with oversampling for capturing extreme event predictions including Halloween, Christmas etc. Developed a explainable framework using feature coefficients for forecasting prediction• Developed curated features for specific cafes using OSM (open street Map ) and change point detection techniques (Ruptures) to improve the forecasting accuracy by 100 basis points. Analyzed the Fourier transformation of time series signals in the linear regression to capture the annual seasonality.• Utilized Kalman smoothening approach to treat the outlier and to denoise the trend variables. Leveraged event based incremental demand analysis from PredictHQ to select statistically significant features . Experimented Bayesian probabilistic updation of model coefficient for time series with very less data points for better accuracy Show less

• Spearheaded a solution based on Image Analytics for a insurance major to help underwriters perform better risk identification of man made structure from aerial images. Implemented U-net to perform semantic segmentation of aerial images on a real time basis for detecting roads,buildings, parking spaces , dense and sparse vegetation from 466 true color images using Keras framework. Used NDVI on IR images to identify the natural and man made structures from 100 IR images. Annotated the… Show more • Spearheaded a solution based on Image Analytics for a insurance major to help underwriters perform better risk identification of man made structure from aerial images. Implemented U-net to perform semantic segmentation of aerial images on a real time basis for detecting roads,buildings, parking spaces , dense and sparse vegetation from 466 true color images using Keras framework. Used NDVI on IR images to identify the natural and man made structures from 100 IR images. Annotated the images using latest techniques and introduced image augmentation techniques in terms brightness, contrast orientation to improve the learnability of the model• Developed a python package to improve the of performance mainstream classification models by creating new automated features using association based rule mining (Apriori Package) with feature binning and discretization. Show less