Sai Vikas Reddy Yeddulamala

Architected a RAG-based, multi-agent customer-support chatbot using LangChain, prompt-engineering, and LoRA-fine-tuned LLMs on 200 K feedback records, boosted NPS by 12% and 15% faster resolution via A/B tests.

Engineered a real-time Azure fraud detection pipeline using an ensemble of Autoencoders for anomaly detection and LSTM for sequential analysis, cutting fraudulent transactions by 15% and transaction latency by 20%. Containerized microservices on Docker and tracked pipelines with MLflow for CI/CD.

Developed ARIMA and Prophet time-series features in Snowflake and SQL to extract seasonality and risk features by improving forecast reliability 5%, then trained LightGBM (churn) and XGBoost (loan default) gradient boosting models, saving $800k in revenue and reducing churn by 10%, risky approvals by 8%, and presented SHAP based insights in Power BI.

Partnered with data engineers to build PySpark EDA pipelines and a reusable feature store on Databricks, integrating Azure CI/CD with MLflow monitoring, accelerating analysis by 15%, cut model refresh cycles by 30% and maintained 99% uptime.

Assisted in building transfer-learning machine learning classifiers with ensemble attention to score personalized customer offers; exposed real-time prediction using GraphQL API's, boosting campaign conversions 8% in A/B tests.

Streamlined SQL based ETL workflows with incremental loads and data-quality checks, reducing processing time by 20% and delivering customer-transaction data to Power BI dashboards for daily-decision making.