I build high-performance systems designed to scale. As a Research Assistant in Prof. Peter Baumann's group at Constructor University I work on scalable geospatial inference for Earth observation, from the data pipeline to the hardware it runs on. Before that: backend engineering in Java and Python.
Rati Vardiashvili, Dimitar Mishev & Peter Baumann
proceedings of the Intelligent Systems Conference 2026 (IntelliSys 2026)
Deep-learning inference framework and a CLI tool able to efficiently analyze scenes that are orders of magnitude larger than available memory. Optimizes for both budget hardware and high-end workstations.
Working full-time as a Research Assistant in Prof. Dr. Peter Baumann's research group. Focused on scientific research and the development of high-performance geospatial data systems.
Engineered Java-based integration solutions to automate interactions with the Georgian Revenue Service. Streamlined data exchange processes and collaborated with cross-functional teams to integrate secure APIs.
Developed an internal security system using Raspberry Pi clusters. Administered the company's main web platform and received in-depth training on Java EE backend architectures.
My Bachelor's thesis: A high-performance inference pipeline for gigapixel satellite imagery. Solves 'grid artifacts' using seamless reconstruction and dynamic memory management.
A robust FUSE filesystem that mounts iCloud Drive on Linux. Features optimistic syncing, sparse caching for large files, and conflict resolution.
Thesis grew into GSIP, published at IntelliSys 2026.
