Rutwik Belsare

Auto Email / Ticket Categorizer

Built a lightweight NLP classifier that reads incoming support tickets (subject + body) and automatically routes them to the correct department — Billing, Technical, HR, or General. • Engineered a TF-IDF + Logistic Regression pipeline achieving 83% classification accuracy, with text preprocessing and precision/recall/F1 evaluation across all categories. • Added a confidence-based human-review fallback for low-certainty predictions and a keyword-driven priority tagger to flag urgent tickets in real time.

Document Q&A System using RAG

Built a Retrieval-Augmented Generation (RAG) pipeline enabling users to upload documents (PDF/text) and ask natural language questions, retrieving context-grounded answers. • Implemented document chunking and embedding using Sentence Transformers; stored and queried vector representations using FAISS for efficient semantic search. • Integrated a pre-trained language model via Hugging Face to generate accurate, context-aware responses, reducing hallucination compared to vanilla LLM prompting.