Designed and developed QuIT, a B+-tree with a fast leaf access optimization, that achieves up to 3.2× faster ingestion for near-sorted workloads. It also reduces the memory footprint by up to 28%, resulting in 1.36× faster range lookup queries. Implemented a concurrent version of the tree using atomic version locks.

Worked on a hardware-based solution that allows memory or storage components to perform on-the-fly transparent data transformation and improves performance by reducing data movement.

Created a live demo for Relational Memory using specialized hardware (Xilinx Zynq ZCU102) for the on-the-fly transformation of row tables into column groups at query time. This work won the Best Demo Paper Award at VLDB 2023.

Created an indexing performance benchmark for near sorted data.

Developed a Horovod extension that supports various compression methods for distributed data-parallel deep-learning training. GRACE integrates communication methods provided by MPI and NCCL. Implemented and compared various compression, communication, and memory compensation methods from literature using PyTorch and TensorFlow (v1/v2). The experiments were performed on NVIDIA V100 GPUs, and the results were visualized with WandB.

Created features that automate integration with various cloud computing providers and backup checkpoints for cloud storage.

Developed features on the company's main product, and deployed progressively with feature flags. Resolved issues, fixed bugs, and participated in team meetings. Automated the procedure that creates SEO-friendly sitemaps for clients.

Worked as a Java Backend Engineer at the Hellenic Ministry of National Defence. Developed web apps, maintained features, and implemented new ones used by the Army Control & Information Management System.

Developed an application that downloads and manages company reports using SOAP web services.