Tunjay Gafarli

Compiler & GPU Engineer | CENG Undergrad @ Türkiye-Azerbaijan University

prof_pic.jpg

Computer Engineering Dept.

Türkiye-Azerbaijan University

Baku, Azerbaijan

I am a Computer Engineering undergraduate at Türkiye-Azerbaijan University, currently serving as a Teaching Assistant for the Computer Laboratory. My research interests lie in systems infrastructure, specifically compiler design, deterministic GPU execution, and memory-safe Edge AI.

I am the creator of Vyne, a systems programming language designed for numerical computing. Vyne compiles a single AST to multiple backends, including x86, CUDA PTX for GPU training, and bare-metal AVR/ESP32 microcontrollers. By utilizing region-based memory management, Vyne enforces strict compile-time memory bounds and deterministic execution, making it possible to train and deploy neural networks on highly constrained hardware.

Beyond systems programming, I have four years of experience in music production. This background drives my interest in real-time Digital Signal Processing (DSP) and building low-latency audio hardware.

I am always open to discussing compilers, CUDA, or embedded systems. Feel free to reach out via email.

news

Nov 17, 2026 Published a hardware benchmark comparing Vyne against TensorFlow Lite for Microcontrollers (TFLM) on a classic ESP32—achieving a 23.4× latency speedup via native Float32 lowering, eliminating linked arena overhead, and breaking down where TFLM still wins. Read the full article here.
Nov 13, 2026 Published a technical deep dive on how the Vyne compiler reduced neural network VRAM usage by 94% and achieved bit-exact deterministic GPU training. Read the full article here.

latest posts

selected publications