Sparsh Kumar

I've built machine learning systems for fraud detection, text correction, and grounded question answering. My credit card fraud pipeline achieved 0.97 ROC-AUC and 92% fraud recall on 284,807 real transactions using SMOTE and XGBoost.

I've developed a hybrid AI-powered text correction system using SymSpell, TextBlob, and a fine-tuned T5 Transformer, reducing Word Error Rate by 27%. I also built a YouTube transcript chatbot with LangChain, FAISS, Google Gemini-1.5-Pro, and Streamlit to retrieve grounded answers from video content.

In Flipkart's Gridlock Hackathon 2.0, I ranked among the top 3% of 35,000+ participants and built an XGBoost traffic forecasting model with an R2 of 0.90. At IIT Kharagpur's General Championship Data Analytics 2025, I helped optimize advertising investment, increasing ROI from 107% to 128% and profit from 7% to 28%.

I'm pursuing a dual degree in Biotechnology and Biochemical Engineering at IIT Kharagpur, with a micro-specialization in Artificial Intelligence and Applications. I bring strong foundations in data structures, algorithms, statistical analysis, and applied machine learning.