Shayan Samimi Sadeh — Istanbul
Computer Engineering student. I design AI tools, trade algorithms, and gamified economies — then write historical fiction when the systems go quiet.
I'm a Computer Engineering student at Altınbaş University with a background that cuts across disciplines — humanities first, then code. That combination shapes how I think about technology: not just how systems work, but who they affect and why it matters.
I currently intern at FlyRank AI, working daily with large language models — not just using them, but learning how they behave, fail, and improve. Outside of that, I architect projects at the edge of what's possible: Parlance, a no-code visual workspace for multi-agent AI orchestration; LoopLens for agentic loop cost visibility; Claude Batch for pacing prompt workflows without hitting rate limits — and more.
Outside of tech, I write the Liege series — historical fiction about iron, honor, and what people are made of under pressure.
A visual workspace for orchestrating multi-agent Claude AI workflows — no code, no YAML, just a node-link canvas. Connect AI agents, file sources, and output endpoints with edges, then run them with real API calls. The execution engine handles parallel wave execution, three-tier retry escalation, and targeted Overseer feedback that re-runs only the agent responsible — not the entire workflow. BYOK, fully browser-based.
A free, open-source tool that queues prompts to Claude and fires them at a controlled, user-defined pace — so you never hit a rate limit mid-workflow. Built for repetitive batch work: data analysis, document processing, running the same prompt across many inputs. Runs entirely in the browser, bring-your-own-key, nothing ever touches a server.
A cost simulator for agentic AI loops — see per-iteration cost breakdowns before running anything. Models fan-out multi-agent architectures, calculates prompt caching break-even, and compares 13+ models side by side on the same loop configuration. Pure deterministic math, zero API calls required to simulate.
A three-stage autonomous algorithmic trading system built on statistical anomaly detection. Six walk-forward runs. The core engine uses z-score–based deviation analysis, a custom capital allocation model (MSIC), and a portfolio heat limiter capped at 10% concurrent open risk.
Working at a next-gen AI search company, deploying LLMs in real marketing workflows. The role goes beyond using the tools — understanding how they behave across different contexts, where they fail, and how to engineer around it.
Combining a technical CS foundation with independent projects across AI, fintech, and system design. Humanities background (prior to university) shapes how I think about tech's real-world impact.
Philosophy essays on Substack and an ongoing historical fiction series on Medium. Two different modes of thinking on the page.