At AREAL.AI, I develop and maintain production document automation solutions, taking assigned components through annotation, training, evaluation, optimization, and deployment. I trained a signature detection model that achieved approximately 99% recall.
I also integrated OpenAI and Anthropic models into document information extraction workflows, adapting prompts and structured outputs during model migrations. Through model and prompt improvements, I achieved approximately 60–80% cost optimization in LLM processing.
Previously, at IND Software, I worked on a TÜBİTAK-funded drone project for post-earthquake damage assessment, fine-tuning Detectron2 segmentation models. My research work at TÜBİTAK included medical image analysis, and I developed a Raspberry Pi-based digital advertising prototype using face detection and age classification.

