Class: Full-Stack Developer · Subclass: AI-Native Systems · Base: Türkiye 🇹🇷
Solo developer — which mostly means I've personally made, and fixed, every mistake in this stack.
Self-assessed. No saving throw was rolled. Take with a grain of salt.
| Stat | Score | Mod | In practice |
|---|---|---|---|
| 💪 STR — Stubbornness | 17 | +3 | Years into a solo FRPG engine and still not give up. That's dedication and consistency |
| ⚡ DEX — Frontend | 14 | +2 | React Native + Expo on mobile; React, Three.js and GSAP on web |
| 🛡️ CON — Backend Endurance | 15 | +2 | 505 Postgres functions and counting. Someone has to maintain them, and that someone is me |
| 🧠 INT — System Architecture | 16 | +3 | A +99-table PostgreSQL schema with RLS on every table — no exceptions, no "we'll fix it later" |
| 🔮 WIS — AI Engineering | 15 | +2 | LLM pipelines where the AI proposes and the server decides. Learned the hard way why that order matters |
| ✨ CHA — Product Sense | 20 | +5 | The table did say self-assessed. But baking the free/premium split into the architecture itself? That was a natural 20 |
A mobile-first RPG where an AI acts as the Dungeon Master, following D&D 5e rules — built around one idea: the AI is not allowed to touch reality.
flowchart LR
A["🔮 AI<br/><i>suggests</i>"] -->|"_change deltas only"| B["🗄️ Database<br/><i>decides & writes</i>"]
B -->|"validated state"| C["📱 App<br/><i>displays</i>"]
style A fill:#7c3aed,color:#fff
style B fill:#1e293b,color:#fff
style C fill:#0ea5e9,color:#fff
The AI never rolls dice and never edits a stat. It writes proposals (_change fields); only the server validates them and updates the real state (_current). Cheating isn't against the rules — it's structurally impossible. You can't fake a nat 20 when the dice live on the server.
| ⚙️ Stack | React Native 0.81 + Expo 54 · Supabase PostgreSQL + Edge Functions (Deno) · Mistral AI |
| 🗄️ Numbers | 112 tables · 505 functions · 272 RLS policies · server-side 4-phase dice |
| 💰 Modes | Sandbox (lightweight) · FRPG (full D&D experience) — different cost profiles by design |
| 🎯 Status | Pre-launch. Taking my time on purpose — I'd rather ship late than ship with security holes |
🌌 Togu Nexus — a community platform with events, forums and role-based moderation. React, TypeScript, Three.js, GSAP — I wanted it to feel like a place, not a webpage. Supabase Realtime keeps the event board live.
🤬 Turkish Profanity Detection ML — a pipeline for detecting Turkish profanity and slang, including the creatively misspelled kind. I built the labeled dataset myself; Turkish internet slang mutates fast, and off-the-shelf filters don't stand a chance.
🎭 Ozan Engine — a small interactive narrative engine living inside Togu Nexus. Typed, tested, and fun to break.
- Never trust the client — in games or anywhere else. The player doesn't get to roll their own dice.
- Documentation is for future me. He shows up three months later with no memory and a lot of questions.
- No deadline is a feature. Use the freedom to get things right, not just done.
