Kapil Kevlani

Developed data extraction and transformation pipelines, converted large PDF datasets into data chunks and utilized Hugging Face's Embedding models to covert chunks into Embeddings, enhance chatbot comprehension by 40%. Engineered a state-of-the-art RAG pipeline utilizing Llama2 and Pinecone Vector Database which meticulously summarized top results, elevating user satisfaction by an impressive 50% through precise, contextually relevant responses. Led the chatbot's deployment on AWS, coupled with FastAPI, accommodating over 5000 monthly user interactions with peak reliability. Creatively designed the user interface utilizing CSS and HTML, achieving a 30% improvement in user engagement. Led AWS Glue ETL pipeline development, transforming data between AWS RDS and S3, enhancing data processing efficiency by 40%. Extracted unstructured data using PySparks from AWS RDS, transformed the data into structured and required format using AWS Glue, Pandas, NumPy and loaded the data into AWS S3. Managed ETL workflows, processing 35,000 records daily with 100% accuracy, and optimized data storage on AWS S3 using fixed-width files. Utilized a broad technology stack, including AWS services, Apache Spark, Python, PostgreSQL, SQL Server, and DB2, showcasing technical adaptability and expertise. Spearheaded an NLP sentiment analysis project, classifying 20,000+ medicine reviews, boosting customer insight and strategy. Pioneered data ingestion, EDA and transformation pipelines to extract data from PostgreSQL and transformed data utilizing Pandas, TfidfVectorizer. Employed Scikit-Learn's models like Decision Trees, Random Forest, SVM and XGBoost and achieved 92% accuracy with an SVM model on vectorized data, increasing the sentiment analysis tool's predictive power. Developed API to call the classification pipeline using FastAPI framework to enable real-time processing and classification of customer review sentiments. Designed CI/CD pipeline releasing machine learning package and data pip

Build End to End Machine Learning projects utilizing python, pandas, SQL databases, supervised (Linear Regression, Lasso, Decision Trees, Ensemble Learning, SVM), unsupervised (K-means Clustering, DB Scan and PCA) and reinforcement learning algorithms. Created Deep learning architecture like ANN, CNN, RNN to solve complex problems using libraries like Tensorflow, PyTorch etc.