Software engineer focused on cloud infrastructure and serving large language models at scale.
I work at Trendyol Group in Istanbul, where I build and operate cloud infrastructure — most recently GPU-based LLM serving platforms on Kubernetes. Before that I developed the mobile API, web services, and AWS infrastructure behind Dolap.com. I write about the things I break and fix along the way: AWS, caching, and inference serving.
Distributing traffic across a large language model running on 20 GPU servers is a harder problem than round-robin. A walk through prefix-aware routing with llm-d, vLLM, and LMCache.
Restricting an SQS-triggered Lambda to daytime windows using CloudWatch Events and UpdateEventSourceMapping.
Why "I'll just buy my own servers" is usually a worse deal than it looks: serverless architecture, Lambda, DynamoDB, and CloudFront costs.
Varnish Cache from the ground up: VCL configuration, backend routing, round-robin load balancing, cache purging, and performance numbers.
Open source on GitHub — 60+ repositories.
Automate switching AWS Lambda functions on and off with Boto3.
Turkish stemmer plugin for Elasticsearch 2.3.x.
Working .ebextensions configurations for AWS Elastic Beanstalk (nginx).
Scalate + Hibernate + MySQL + Spring MVC/Spring Boot sample application.