I'm a senior AI engineer and PhD researcher. For six-plus years I've designed and shipped production AI systems — owning architecture, system design, and tech-stack decisions across health tech, insurance, legal tech, and aerospace. Now I'm focused on research — foundation models, continual learning, and neuro-symbolic AI.
Recent
- Jun 2026Won the DocVQA challenge at ICDAR 2026.
- Sep 2025Started a PhD in Information Systems at METU.
- Oct 2024Completed an MSc in Data Informatics at METU.
Writing
MSc Thesis — Knowledge Graph Augmented Multi-Hop Question Answering Using Large Language Models. METU, 2024. Open archive
- Perceive-Reason-Code: Active Perception for Document VQAJul 2026How a code REPL plus on-demand VLM perception let an open 27B win a document-QA challenge, and which parts carry it.
- Fine-tuning vs. PromptingFeb 2025When adapting weights beats prompting, and what it costs.
- Is Everything an Abstraction?Jan 2025Physics, Kant, and the layers we build to grasp reality.
Experience
Senior AI Engineer — Independent Contractor
AI Software Engineer — Peppy Health
Senior Software Engineer — Tractable
Software Development Engineer — Amazon Web Services
Projects
Feb 2026
Abstract Reasoning (epiq)
An LLM agent that solves ARC-AGI puzzles by implicit program synthesis in a Python REPL, with holdout verification. Code
Jun 2025
Fine-tuning vs. Prompting
Prompt optimization matches LoRA fine-tuning at lower cost on entity–relation extraction. Code · Post
Apr 2025
RL for Multi-Hop QA
GRPO-trained Llama-3.1-8B learns agentic retrieval — answer F1 0.30 → 0.48 on MuSiQue. Code