V V Chakravarthi

Worked on building a Gen AI service engine that utilizes the Lite LLM (Large Language Model) for generating responses, integrated with Azure Chat GPT 4.0 model API keys. This engine is designed specifically for extracting tabular content from Statement of Values worksheets. Developed benchmark pipeline to evaluate the model performance for both extraction and data normalization. Built a service that employs an XGBoost model and a LayoutLM transformer model for document classification, along with the Flair framework for extracting relevant entities.

Lead Engineer for BYJU’S QnA search engine. Management of Search Success rate, Search trends and Question hits traffic. Student Sentiment Analysis for BYJU’S QnA product using NLP techniques and classification algorithms. Autograding project involved validating the handwritten answers for math equations printed on a worksheet. Created a synthetic dataset of 10000 images for equation object detection task using YOLOv5. Created the Backend API using Flask and Python, deployed on Amazon EC2.

Designing API datapoints that delivers a high touch, effective and targeted Customer Engagement capability for Client Mister. Integrating Google Tag Manager Data Layer with the product front end and pushing the useful Customer e-commerce data to Braze platform. Analysing and Segmenting the data to our needs in Braze and orchestrating Campaigns and Canvas. Tracking Customer Analytical events with the help of Google Analytics(GA4) through GTM and Funnel Analysis of User Churn and Retention.

Preparing Data by building ETL pipelines from Azure BigData to Azure Data Explorer for Client Microsoft. Building Power BI application for Enterprise, Consumer Microsoft Groups data according to Business needs. Leveraging Machine Learning techniques for forecasting trends of Active Users. Anomaly detection in Consumer and Enterprise groups data using Power BI Python scripting.

Natural Language Processing: Task: Text Classification. Used TF-IDF vectors and word embeddings in feature extraction on Skicit’s 20 Newsgroups data set. Build Naive Bayes and SVM ML Models that are giving accuracy of nearly 90percentile and 87percentile.