Sidharth Sankar Sabat

Developed a GenAI-powered corporate Q&A system using Meta LLaMA and Google Gemini, enabling natural-language insights across structured and unstructured data with automated plot generation. Built a GPT-Neo–based customer support chatbot to automate enterprise workflows, query resolution, and ticket classification. Created a multimodal image captioning system combining CNN encoders with Transformer decoders; trained on Flickr8k and evaluated using BLEU, CIDEr, and SPICE. Designed an image captioning pipeline leveraging Vision Transformer encoders and Transformer-based decoders for improved context retention and caption quality. Implemented a deep learning U-Net pipeline for rooftop segmentation (academic project), including data augmentation, loss optimization, and evaluation on satellite imagery datasets. Built an NLP entity and relation extraction pipeline using spaCy and Hugging Face Transformers, with output visualized via a knowledge graph. Developed an LLM-based data insights tool using LangChain, enabling natural-language querying over SQL databases (Text-to-SQL) and unstructured document search (RAG). Integrated LangChain agents, tools, memory, vectorstores (FAISS/Chroma), and prompt templates into multiple coursework projects to handle chaining, summarization, retrieval, and instruction following. Designed multiple end-to-end ML pipelines covering data cleaning, feature engineering, model selection, hyperparameter tuning, and evaluation using Python, PyTorch, and Scikit-learn. Completed advanced coursework in Machine Learning, Deep Learning, NLP, Cloud Computing, Big Data Systems, Time-Series Forecasting, and Statistical Modeling. Presented research-based insights using Tableau, Matplotlib, Seaborn, and Plotly for academic case studies and project reviews. Worked in collaborative teams using agile practices, contributing to design discussions, model experimentation, documentation, and deployment-oriented presentations.