ROSHAN VARGHESE

I've built and deployed AI systems across retrieval, multi-agent orchestration, and computer vision, including RAG applications for agriculture and enterprise workflows.

At Codexoho Digitals, I fine-tuned and deployed an OpenCV facial-recognition attendance pipeline using FastAPI, Next.js, React, PostgreSQL, Docker, VPS infrastructure, and RTSP camera feeds. I piloted it on two production cameras and investigated GPU-efficient scaling for a 40-camera rollout.

At KOTS Daily Services, I designed a three-agent LangChain and Google ADK workflow and built a FAISS- and Gemini-based RAG pipeline over PDFs and chat history. Earlier at Deloitte Consulting USI, I validated 184 ETL jobs supporting 80+ SAP BI reports and reduced manual QA effort by 60% through defect root-cause analysis and an automated PDF validation proof of concept.

I'm especially interested in grounded LLM applications: my projects include KrishiSahayak, a cited agricultural-schemes chatbot with ChromaDB and web-search fallback, and a four-agent agricultural system combining RAG, forecasting, and pest detection.