Manidedeepya Chennapragada

Developed and fine-tuned large language models using PyTorch and LLaMA 2/3 to enhance generative ad copy, boosting click-through rates by 11%. Built and deployed vision diffusion models for image background generation and animated video creatives, increasing ad engagement by 13% and accelerating ad delivery by 60%. Conducted research and prompt-tuning experiments improving ad copy relevance by 9%. Engineered real-time ad recommendation models using deep neural networks and reinforcement learning, contributing to a 7.6% conversion rate lift. Designed scalable inference pipelines with ONNX, TorchScript, achieving sub-100ms latency supporting 10M+ daily requests. Led canary deployments and A/B tests validating 5%+ uplift at production scale. Integrated multi-modal pipelines increasing platform engagement by 8%. Developed real-time scalable data pipelines and monitoring dashboards. Collaborated cross-functionally to implement AI watermarks and safety filters, reducing unsafe events by 35%.

Developed and deployed credit risk models reducing loan default rates by 23%. Designed NLP pipelines for financial document classification improving compliance accuracy by 30%. Built real-time fraud detection systems processing millions of transactions daily, reducing fraud loss by 40%. Automated ML lifecycle cutting deployment time from 2 weeks to under 48 hours. Developed explainable AI workflows for transparency and compliance with regulatory guidelines. Led AI-powered chatbot deployment achieving 90% query resolution accuracy. Collaborated in Agile teams delivering scalable AI solutions supporting $50M+ portfolios. Created predictive analytics dashboards improving executive decision-making. Integrated ML models with enterprise systems reducing manual reporting. Spearheaded Responsible AI initiatives to audit and reduce bias in lending models.