At Tensor Yapay Zeka ve Bilişim Teknolojileri A.Ş., I shipped Hukuk AI, a legal assistant over 4.7 million Turkish court decisions and legislation. Its hybrid keyword and vector search powers the paid tier of an app with 5,000+ users.
I adapted a Turkish transformer language model to legal text with masked-language pretraining, then fine-tuned the dense retriever. I also built the retrieval benchmark used to compare models, with gold documents hidden among 5,000 distractors and recall-at-K scoring.
I built tool-calling agents and MCP servers for the legal assistant and five apps, with confirmation before changes, audit logs, and limits. I also worked on Mobil UDF Editör, an iOS app for opening, editing, e-signing, and sharing UYAP court documents, launched in January 2026.
My earlier work includes information extraction and document classification at Yapı Kredi Teknoloji, vision models for retinal OCT and Alzheimer’s diagnosis at TÜBİTAK, and a real-time vehicle-camera preprocessing tool at HAVELSAN. I’m pursuing an MSc in Computer Sciences at Istanbul Technical University and researching joint-embedding predictive architectures.

