George David Tsitlauri
From hardware to AI.
Who I Am
Research Lab
Member of the CACS Lab (Computer Architecture, Compilers & System Security), University of Thessaly.
Master of Science
Artificial Intelligence & Applications, University of Thessaly, Greece.
Bachelor of Science
(Data Management & AI)
Informatics & Telecommunications, University of Thessaly, Greece.
What Drives Me
I enjoy research, exploring new ideas, and learning across the stack — from hardware and computer architecture to AI — so I can understand and build systems end to end.
“Time and Patience.”
— Leo Tolstoy
Research Projects
Crypto3DStackCPU (Preliminary PhD Research)
A processor that runs only encrypted, signed programs on top of a four-layer 3D-stacked memory. Every transfer between the memory layers is cryptographically signed, so tampering, replayed data and physical fault attacks are detected, and the chip locks itself and erases its keys when it is attacked.
Supervisor: Prof. G. Dimitriou
CUDA-zkML (Preliminary PhD Research)
A system that lets a service prove that an answer really came from the neural network it claims, without revealing the model or the user’s data. The short cryptographic proofs are generated on a graphics card and can be checked in milliseconds, even by a smart contract.
Supervisor: Prof. G. Dimitriou
TELEMACHUS (MSc Thesis)
A method that teaches an AI agent when it knows enough to act and when it should gather more information, think longer or ask a human for help. The maximum error rate is set in advance, and the agent is guaranteed, with high probability, to stay within it while spending as little time, computation and human help as possible.
Supervisor: Prof. G. Dimitriou
Academic Projects
CryptoCPU (BSc Thesis)
A MIPS-style processor on FPGA with an AES-128 decryption stage inside the pipeline, so it runs only encrypted programs and no instruction ever appears in plain form in memory.
Supervisor: Prof. G. Dimitriou
NEXUS
Compiler for a small C-like language with register allocation and SSA, CPU simulators from single-cycle to out-of-order (Tomasulo), multicore and GPU models, and a pipelined MIPS processor in Verilog.
CCC IR Optimizer
Optimizer in C for the CCC high-level-synthesis toolchain: region-based and iterative data-flow analysis, constant propagation, loop-invariant code motion, common-subexpression and dead-code elimination.
Parallel AES
AES-128 encryption on NVIDIA GPUs in CUDA C++ (ECB and CTR modes), with one GPU thread per 16-byte block, verified against FIPS-197 and OpenSSL and reaching 12–15 GB/s.
GPBFT
Byzantine fault-tolerant consensus protocol that groups nodes and aggregates their votes to reduce communication, implemented in Python and evaluated against PBFT in the SymBChainSim blockchain simulator.
Web Projects
Technical Skills
Programming
AI / Data
Hardware
Web
Dev
Spoken Languages
Greek
Native
Russian
Native
English
ECCE – University of Michigan