This study presents an efficient Extract, Transform, Load (ETL) process for synchronizing data between NoSQL time series databases and relational data warehouses in IoT environments, addressing challenges of maintaining data consistency while supporting real-time and historical analytics. Our methodology uses Azure CosmosDB for NoSQL storage and Azure Synapse for warehousing. The ETL process is optimized for high-velocity, high-volume IoT data, focusing on data modeling, transformation, and loading. Key optimizations include incremental loading, parallel processing, and data compression. Query times improved by over 99%. The system manages out-of-order data arrival through staging, windowing, and merge operations. This approach balances real-time accessibility with powerful analytics, enhancing IoT data processing across domains beyond smart homes.