Re-engineered the end to end Machine Learning pipeline architecture for a critical OCR project, successfully slashing total model training time by 50%. Conducted model optimization research that led to a 20% improvement in prediction accuracy and enhanced inference time efficiency by 6% on average. Developed key data efficiency tools to support the pipeline, including a semi supervised labeling system (30% less annotation time) and automated scraping tools (40% faster data collection). Spearheaded the research and development of an ERP integrated Whatsapp chatbot PoC, enabling natural language querying of business data.