I build AI apps and systems — and I run them in production.
Self-taught developer, building software since 2021. For the past three years I have built with AI-assisted workflows — directing coding agents, verifying their output, and holding it to production standards. My proof is not a diploma; it is a live product, below, that real customers can pay for and use today.
A credits-based AI platform I designed, built, and operate alone. A customer buys a prepaid key once and can use 48 AI models — Claude, GPT, Gemini, DeepSeek, and more — across chat apps, coding tools, and terminals. Every part below is a separate program I shipped.
The money brain, on Cloudflare Workers. Per-model credit pricing computed from real upstream costs, a hard spend cap on every key, automatic routing between sellers with failover, and real-time balance tracking. Built so a key can never spend more than it was sold for.
Whop checkout, crypto, and QR Ph. Keys are created or refilled automatically when a payment is confirmed, and delivered by email. A phone-based inbox (SMS gateway + notification listeners) matches incoming payments by amount and time.
A Tauri/Rust desktop widget showing the live credit balance, connection status, and per-model usage estimates. Ships as a signed-off installer on the site.
A one-key setup app: paste a key and it configures Codex, Claude Code, OpenCode, and Hermes — with backups, merge-never-replace writes, and read-back proof before claiming success.
A fork of OpenCode with ZOLT baked in as the only provider, a live per-step credit meter, and session budget caps.
A full rebrand and rebuild of the Cherry Studio app (AGPL) as ZOLT's own AI app — new identity, credits-aware defaults, and ZOLT-only plumbing.
A Cursor companion app (fork of cursor-byok) whose model list syncs itself from the live catalog, plus VS Code/Cursor extensions that show the real credit balance in the status bar.
Telegram and Facebook bots wired to the gateway's admin API, and a web support inbox with agent claiming and quick replies. I am the whole support team.
The same 20 small jobs — writing, questions, code, logic — sent to all 48 models on my own gateway, every answer scored by rule and the code answers actually run. Three near-free models scored a perfect 20/20 (MiMo V2.6 Flash, GLM 5.3 Flash, GLM 5.3 FlashX) while the famous flagships landed at 18/20. Total measured cost of the whole run: 41.91 credits.
Full results table →Engaged for AI red teaming work in 2026. Scope and methods are covered by an NDA and are not shown here.
Server-side detection for live multiplayer servers, in Lua and C#: teleport, speed, and injection exploits.
A C# Blazor crosshair overlay desktop app with real-time rendering and custom profiles.
A Windows optimization suite (OS tuning, registry automation, BIOS guidance) distributed to end users, and custom peripheral firmware in C/C++ for RP2040 microcontrollers.
multi-model API gateways, usage metering and pricing, model routing and failover, AI agent tooling (Cursor, Windsurf, Qodo), prompt design, output verification
JavaScript / TypeScript, Python, C#, C/C++, Lua, HTML/CSS, Bash
Node.js, Cloudflare Workers, REST APIs, React, Next.js, Blazor, Tailwind CSS
Tauri (Rust), Windows internals, VS Code / Cursor extensions
vulnerability analysis, anti-cheat engineering, reverse engineering, secure coding, risk assessment
Playwright, Selenium, custom bots (Telegram, Facebook), payment integrations (Whop, crypto, QR Ph), Git, CI/CD
Fastest reply is email — I read everything myself.
Silver Angelo Bonto · Quezon City, Metro Manila, Philippines