Pulkit Sharma

I've built an RLHF preference-training pipeline from scratch, including reward modelling, PPO fine-tuning, LoRA/PEFT, and a Gradio annotation interface that collected roughly 1,200 labelled preference pairs. I also built DocuAgent, an eight-agent RAG system for document retrieval that supports 50+ concurrent queries with sub-500ms retrieval latency.

During my Data Science internship at Coratia Technologies, I developed an LSTM model to predict machine failure rates and identified an approximately 12% cost-saving opportunity through ROI modelling. At Search Education, I analyzed 15,000+ CRM records and automated 5+ Excel and Power BI MIS pipelines, reducing manual reporting effort by about 30%.

I'm completing a B.Tech in AI & Data Science while gaining hands-on experience in data engineering at HCL Technologies and Velora Enterprise. My work spans Python automation, SQL, ETL, model evaluation, APIs, dashboards, and deployed ML products.