Nisarg Choudhary

Source Code

• Built an AI platform for retail demand forecasting using the M5 Forecasting dataset, integrating ETL pipelines, a MySQL star-schema warehouse, business analytics, and forecasting workflows over 58M+ retail sales records.
• Implemented and benchmarked 10 forecasting models across classical, deep learning, and transformer approaches using a common training and evaluation pipeline with experiment tracking and checkpointing.
• Used Google Gemini to generate natural-language summaries of forecasting results, trend analysis, seasonality, and statistical metrics, exporting reports in Markdown and HTML formats.
• Built a Retrieval-Augmented Generation (RAG) pipeline using SentenceTransformers and FAISS to answer forecasting and inventory-related questions from project documentation while citing the retrieved context.

Source Code

• Built a machine learning pipeline to predict telecom customer churn using historical usage, recharge, and revenue data.
• Performed preprocessing, feature engineering, churn label generation, and class imbalance handling using SMOTE.
• Trained Random Forest and Logistic Regression models achieving approximately 0.83 recall, and deployed an interactive Streamlit dashboard for customer risk segmentation.

Source Code

• Developed a neural language model for next-word prediction using the Shakespeare Hamlet dataset.
• Implemented and compared SimpleRNN, LSTM, and GRU architectures in TensorFlow/Keras after preprocessing and tokenizing the text corpus.
• Achieved approximately 83% training accuracy and deployed an interactive Streamlit application for real-time Shakespeare-style text generation.

Source Code

• Developed a sentiment classification application using the IMDB movie reviews dataset.
• Implemented a SimpleRNN-based deep learning model with text preprocessing, tokenization, sequence padding, and vocabulary encoding using TensorFlow/Keras.
• Developed an interactive Streamlit application for real-time sentiment prediction of user-provided reviews.