Shashank Kamble

At Rutgers University, I’m pursuing an M.S. in IT & Analytics with an AI concentration while building projects in retrieval-augmented generation and LLM evaluation. For a housing recommendation system, I benchmarked five retrieval pipelines and improved Recall@5 from 0.52 to 0.69 with SPLADE-v3 and Qwen3-Reranker-4B. I also built category-aware FAISS and SPLADE indexes over more than 46,000 structured chunks. Grounded generation with Qwen2.5-32B-Instruct achieved 76.43% answer accuracy and a 2.86% hallucination rate. In my Rutgers Business School Analytics Practicum engagement with Moxie Salon & Beauty Bar, I built a data pipeline integrating POS, payroll, and accounting data across three locations. The analysis surfaced a roughly 12% labor cost overrun at one location, and I’m developing a demand-to-labor forecasting model. Earlier, as a Research & Programming Intern at All India Radio (AIR), I built Python scripts to analyze broadcast and audience datasets and automated a reporting workflow. I’ve also built a full-stack health analytics platform using Python and PostgreSQL, and fine-tuned and evaluated open-source LLMs on a psychotherapy dataset.