Arman Kayhan

I build and study production AI systems.

I’m Ataberk Arman Kayhan — an aerospace engineer by training, AI researcher by background (MSc.), and the co-CEO and co-founder of Co-one.

My work has moved across active learning, IoT systems, enterprise AI evaluation, and reliability infrastructure for production AI agents.

This site is where I publish occasional research notes, technical memos, and experiments on AI systems, inference, evaluation, and reliability.

Current Focus:
• Memory-time tradeoffs in LLM inference
• Evaluation methods for production AI agents
• Reliability testing for enterprise AI systems

Selected Work

Co-one AI LAB ↗

Enterprise AI agent testing and monitoring platform for evaluating grounding, hallucination, safety, and model behavior in LLM applications.

Pickable AI ↗

A product for comparing multiple AI model outputs side by side.

Smart embedded bathroom control panel ↗

IoT-connected smart panel and data systems for connected hygiene products.

CanSat / APIS ↗

A university aerospace project involving a can-sized satellite that collected sensor readings during descent.

Featured Memos
MAY 2026 // SYSTEMSEmpirical SRAM Gating and Semantic GhostsA V2 architecture for KV cache optimization that replaces crude token dropping with empirical compression, squashing low-attention tokens into dense semantic ghosts.FEB 2026 // SYSTEMSMemory-Time Tradeoffs in LLM Inference: A Practical KV-Cache ExperimentAn architectural KV-cache experiment that retained 16.5% of the baseline cache footprint and increased measured throughput by 27% in a high-concurrency Qwen 0.5B stress test, while revealing limitations of blunt token eviction.JAN 2026 // SYSTEMSA Fun Experiment: The Nitro-Protocol (CyLU)Just a project I was tinkering with because I think physics is cool. A custom activation approximating the Brachistochrone curve for LLM training.View all notes →