İbrahim Karamavuş

I've built and deployed two live AI products end to end: a multi-agent LLM/VLM content platform and an agentic audio generation system. I own their architecture, model pipelines, evaluation, web applications, and deployment.

At Codeway Studios, I built diffusion-based image generation and editing systems for Retake, a consumer AI photo application used by millions. I also reduced personalization training costs by up to 80% through GPU optimization, batching, quantization, model compression, and memory-efficient execution.

My work spans large-scale image data pipelines, LLM fine-tuning, distributed PyTorch training, evaluation harnesses, reliability safeguards, and low-latency GCP deployment. Earlier, I developed YOLO-based UAV perception systems and real-time automotive ECU software.