umutcagand · Systems Developer
I build things close to the machine.
Local AI, firmware reverse engineering, and low-level systems experiments with working code, reproducible notes, and measured hardware behavior.
- AMD XDNA1
- AIE2 kernels
- Local AI
- Firmware RE
- Low-level Linux
Selected projects
All projects →Hawk Point NPU LLM
ActiveRunning small transformer models directly on first-generation AMD XDNA NPUs with MLIR-AIE/IRON, plus an OpenAI-compatible local API.
Python · C++ / AIE2 · MLIR-AIE · XRT · OpenAI API
Maintainer of the Samsung Galaxy M20 (m20lte, Exynos 7904) for LineageOS 22.2 (Android 15), now an official device with weekly builds: device trees, a 4.4 kernel with CIP updates, proprietary blobs, and a camera HAL patch for the ultra-wide lens.
Android · LineageOS · Linux Kernel · Exynos · Samsung
Reverse engineering the BES1306 firmware, OTA container, BLE/GATT protocol, and update trust boundaries of the JBL Wave Buds 2, with reproducible unpacking and repacking tools.
Python · BES1306 · Firmware · Reverse Engineering · BLE · Security
Evidence-driven static analysis of the Prestige Client loader and injection chain, documenting unsigned remote payload execution, process injection, and the boundary between proven findings and unverified claims.
Reverse Engineering · Windows PE · YARA · Malware Analysis · Security Research
XDNA NPU Miner
ExperimentalAn early-stage research project exploring whether suitable cryptocurrency mining kernels can be mapped honestly and measurably onto AMD XDNA1 NPUs.
Python · AIE2 · MLIR-AIE · XRT · Ryzen AI
Reverse engineering and reproducible tooling for the Samsung Odyssey G5 G55C firmware, including a minimal patch that exposes an existing hidden factory page.
Python · SPARC V8 · 8051 · Firmware · Reverse Engineering
Latest public activity
GitHub ↗Engineering notes
All notes →- Validating work on AMD XDNA1XDNA1
A practical validation pattern for separating NPU execution from CPU orchestration and checking correctness before performance claims.
- A reproducible firmware research workflowFirmware
How I structure firmware extraction and reverse-engineering notes so every conclusion can be traced back to an artifact.
- Where local AI accelerator boundaries really areLocal AI
A systems view of local inference: tokenization, orchestration, memory movement, kernels, and why an NPU result is more than a single throughput number.