At Thinkbiz Technology Pvt Ltd, I built a LightGBM demand forecasting system for a Japanese retail client, predicting product dispatch quantities from distribution centers to stores over a 20-day horizon. I raised forecast accuracy from 80% to 86–88% on sparse, intermittent demand data.
I turned those forecasts into fixed and spot truck plans, helping the client cut 5–10 trucks per week. I also built a serverless AWS ML pipeline for batched training and scheduled inference.
For a Japanese construction client, I built an advanced RAG pipeline that combines Japanese OCR, LLM-generated request checkpoints with human approval, and product retrieval. It also generates draft quotations with product and labour cost breakdowns.
My projects include HubScout, a LangGraph multi-agent system that verifies models, datasets, and papers, and IntelliDoc, a multilingual RAG Q&A system. I’ve also developed AI interview preparation and automated attendance platforms, and I’m an AWS Certified Cloud Practitioner.
