+ Leading a project to commercialize homomorphic encryption.

+ Worked on a CNN-based indoor positioning system that fuses several signals.+ Developed the entire product chain: a mobile app to collect the signals fortraining, a backend server to gather the data, pytorch training framework for trying out various schemes and parameters, and an on-device CNN trainer.

+ Worked on a cloud-based robot fleet control system for construction sites and an event handler for the robot application logic.+ Productized a vision-based object recognition model for piece-picking tasks. Created gRPC based C library and a web-based management console.

+ Led projects that need coordination across multiple layers. Major projects include: 1) native text index using inverted indexes, with partitioning and CJK language support; 2) B-tree traversal algorithms to enable asynchronous I/O; 3) referential partitioning, a partitioning method that reduces data redundancy using referential constraints; and 4) deferrable constraints, a way to postpone column constraint checks until end of transaction.+ Worked on Query Optimizer Layer, which optimizes… Show more + Led projects that need coordination across multiple layers. Major projects include: 1) native text index using inverted indexes, with partitioning and CJK language support; 2) B-tree traversal algorithms to enable asynchronous I/O; 3) referential partitioning, a partitioning method that reduces data redundancy using referential constraints; and 4) deferrable constraints, a way to postpone column constraint checks until end of transaction.+ Worked on Query Optimizer Layer, which optimizes the query execution tree based on costs. Designed top-down join, which tries out join orders in a top-down manner, instead of the previous bottom-up.+ Worked on Executor Layer, which supports data operations such as table/index scan, sort, join, and aggregation. Optimized string functions, such as substring, trim and replace, by factors of 2 to 10.+ Created and maintained a web-app based issue management system for internal use among fellow developers. Show less

+ Led the project for consolidating over 10 background processes into 4, one of the major features of Tibero 6. Designed the common structure to be used by all processes. Architected the recovery process, one of the 4 background processes, to enable both intra-node and inter-node parallel recovery (patent in progress) for cluster databases.+ Led Recovery Layer, which supports backup, recovery, and replication. Led several redesigns and new features of the layer to increase efficiency… Show more + Led the project for consolidating over 10 background processes into 4, one of the major features of Tibero 6. Designed the common structure to be used by all processes. Architected the recovery process, one of the 4 background processes, to enable both intra-node and inter-node parallel recovery (patent in progress) for cluster databases.+ Led Recovery Layer, which supports backup, recovery, and replication. Led several redesigns and new features of the layer to increase efficiency, robustness and flexibility. Major projects include: 1) Cluster support for active-standby cache replication; 2) recovery manager, an external backup and recovery tool; 3) memory-heavy recovery algorithm to reduce disk I/O by a factor of 20; 4) flashback database, which rolls back the entire DB to a specified point in time; and 5) block change tracker for faster incremental backups.+ Led joint project with IBM to integrate CAPI, IBM’s new hardware interface for SSDs.+ Worked on Cluster Layer, which supports active cache clustering. Stabilized recovery of failed cluster nodes.+ Redesigned and implemented garbage collection for data structure used by global lock management.+ Redesigned and implemented Global View, which allows querying database status across cluster nodes.+ Received Tmax Software Maestro, awarded to the best developer of the year, in Jan 2016. Show less

+ Worked on Data Layer, which supports data structures for tables, indexes, and redo logs used by the database.+ Developed operations for index-oriented table, a hybrid of B-tree index and table.

• Worked in Applied Speech Technology team. Built large-scale language models by merging smaller models. Optimized accuracy by adjusting merging weights based on feature set including error rates and data size. Created robust infrastructure for quickly testing different merging schemes. Dropped failure rates of merge from 2/3 to less than 0.1%.