I built a PDF-constrained conversational agent that answers from uploaded documents, combining dense retrieval and BM25 with reranking. I added a three-layer verification approach to address hallucinations, plus an agentic tool loop and Redis-backed session memory.
For my South Asian Cuisine RAG Assistant, I designed a retrieval pipeline and a 100-question benchmark across eight countries, then evaluated responses with ROUGE, BLEU, and BERTScore. My background also includes testing cloud infrastructure programs at Hitachi Digital and contributing to analytical deliverables as an SAP Intern at Hitachi Vantara.
