Skip to content
View yigaykut's full-sized avatar

Block or report yigaykut

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
yigaykut/README.md

Yiğitcan

I write code that sits between machine learning and systems work. Lately that has meant local vision models driving hardware, GPU kernels, and cryptography.

Some of what I have built

yaykut-cryptology

A cipher where the ciphertext carries a mathematical formula's identity and its parameters instead of plain text. There are three layers. The engine is deterministic and does the encryption. A small neural network sits beside it as an auditor, trying to tell one ciphertext from another, and it never touches the cipher path. The third layer answers questions about the formula corpus.

The repository also holds a constant-time X25519 core written in C, a fuzzer that follows branch coverage through sys.monitoring, a cycle-level timing harness, and a tool that reads the compiler's assembly output to check the constant-time code survived optimisation. There are 29 decision records explaining the reasoning, and the audit document says plainly which claims were measured and which were not.

GPU-Accelerated-Template-Matching

Normalised cross-correlation template matching, written as PyTorch GPU kernels.

Vision-Guided-4-Axis-Robot-Arm-with-Object-Detection

A four-axis arm that finds an object, tracks it and picks it up. A local Qwen vision model does the seeing, an Arduino Uno drives the servos.

Animatronic-Eye-Mechanism

An eye that follows what the camera sees, again with a local Qwen model and an Arduino behind it.

stock-ranking-engine

A screener that ranks US small and mid cap equities against 28 factors, run daily.

What I am interested in

Writing code whose timing does not depend on secrets, and then measuring whether that is actually true rather than assuming it. Fuzzing. Keeping statistical models away from the parts of a system that have to be exactly right.

Pinned Loading

  1. GPU-Accelerated-Template-Matching GPU-Accelerated-Template-Matching Public

    A high-performance Python utility for performing Normalized Cross-Correlation template matching using PyTorch GPU kernels. This module provides a significant speedup over standard CPU-based methods…

    Python 2

  2. Vision-Guided-4-Axis-Robot-Arm-with-Object-Detection Vision-Guided-4-Axis-Robot-Arm-with-Object-Detection Public

    A 4-axis robot arm that tracks and grabs objects using a local Qwen vision model and Arduino Uno.

    Python

  3. Animatronic-Eye-Mechanism Animatronic-Eye-Mechanism Public

    AI powered animatronic eye mechanism using a local Qwen model for real time visual tracking via Arduino

    Python

  4. pri-ai-test pri-ai-test Public

    A character-level transformer language model written from scratch in PyTorch, trained on ten classic books from Project Gutenberg.

    Jupyter Notebook

  5. stock-ranking-engine stock-ranking-engine Public

    Daily multi-factor screener for US small and mid-caps. 28 parameters, encrypted publishing, self-feeding learning loop.

    Python 1