At Amazon Ads, I developed a production ranking system serving 500M+ daily ad impressions at 85ms P99 latency, improving click-through rate by 18%.
I also implemented agentic AI workflows with LangGraph and Model Context Protocol, reducing manual review effort by 40% and shortening review cycles from five days to 18 hours.
For advertising applications, I fine-tuned Llama 3 and Mistral models and built retrieval pipelines across FAISS and Amazon OpenSearch. These efforts improved ROAS prediction accuracy by 15% and Precision@10 by 19%.
Previously at Infosys Finacle, I designed credit risk and fraud detection models for banking clients, achieving 91% precision. I also built cloud-native ML pipelines and automated MLOps workflows that reduced infrastructure costs and model refresh cycles.
