• Refactored the error handling system for the project, providing users more structural and intuitive error messages and making it easier for developers to construct and handle errors.• Implemented background execution pipeline for lazy queries, potentially saving more than 50% of user time on computationally demanding tasks in R, while allowing users to retain full control of their sessions during the computation if needed.• Exported various missing functionalities from Polars to R, such as optimization toggles and streaming sinks for lazy queries.• More details can be found at https://sicheng-pan.github.io/GSoC-2023/

• Implemented approximate queries on Duolingo Metrics, the company’s internal dashboard for querying information about Duolingo users, in order to save 50% of user time and 50% of computation time at the cost of less than 1% uncertainty for most queries.• Implemented caching mechanism for handling common queries based on historical query frequencies, enabling roughly 80% of user time and more than 99% of computation time to be saved at the cost of negligible storage spaces and less than 1% uncertainty for cached queries.• Designed and performed a benchmark to measure the performance of the query approximations, providing substantial insights into the optimal configuration of the cache parameter which maximizes the cache performance through analysis and models based on collected data.