My work tends to begin with questions that are difficult to answer honestly:

• Can an AI coding agent remain powerful without taking control away from the developer?
• Can a product make useful real-time decisions without exporting raw behavioral data?
• Can an anonymity mechanism prove that its complexity buys measurable protection?
• Can groups of people, models, and tools produce decisions that are more verifiable than a single persuasive answer?

I turn those questions into protocols, runtimes, experiments, and products.

An open-source, terminal-native AI coding runtime for real repository work. Aurict combines project-aware context, typed tools, guarded execution, persistent sessions, MCP, LSP, skills, provider-agnostic BYOK routing, and isolated multi-agent workflows.

TypeScript Bun React Ink SQLite MCP Flutter

Website · Repository · Documentation

A privacy-native decision engine that turns live first-party context into one governed, explainable next-best action. Its on-device runtime is designed to keep raw session behavior local while a compact control plane manages models, policy, experiments, and rollout safety.

On-device ML TypeScript Python TFLite Kafka React Native

Product

An anonymity-network research platform written in Rust. MIRAGE uses deterministic experiments, a pinned Tor baseline, explicit attacker models, protocol vectors, and ablation gates to test which traffic-shaping mechanisms provide measurable value—and to remove the ones that do not.

Rust Network privacy Simulation Threat modeling Protocol design

Current boundary: research evidence, not a production anonymity guarantee.

• Local-first and BYOK AI developer tooling
• Multi-agent coordination with explicit authority boundaries
• Privacy-preserving inference and decision systems
• Reproducible security and anonymity research
• Evidence, provenance, and verification infrastructure

• Web: aurict.com
• GitHub: @mikailustuner
• LinkedIn: Mikail Üstüner