Frontend team lead in Istanbul, at HangiKredi. I build interfaces for AI — turning model behaviour into something you can see and reason about.
Lately I've been building developer tools that make the Claude API legible: prompt caching, token economics, agent tool-loops and interactive sandboxes — all in the open.
X-ray your Claude API calls — prompt caching, token composition and cost, visualized live as the response streams. |
Replay a Claude agent's tool-calling loop step by step on a cinematic timeline. |
A/B test prompts side by side — one input, two variants, compared on cost and latency. |
Interactive Claude tool-use sandbox — define tools, mock the responses, watch the agent loop live. |
See a Claude prompt before you ship it — token breakdown, context-window position, cost estimate, prompt-caching boundaries. Pre-flight to claudoscope's post-hoc x-ray. |
|
Long-form posts on each tool: ferhatatagun.com/blog · Suite overview: ferhatatagun.com/tools
Istanbul, Türkiye · open to interesting AI-native frontend work