Turbowork is a one-founder software company in South Korea. Every product — desktop tools for sellers and creators, games, contest entries — is written by AI coding agents inside the Red-Sandwich Harness, where locked specs, hard code limits and an independent red-team auditor decide what ships.
Our core method, published as an open specification on GitHub. AI-generated code can pass a happy-path example while missing edge cases or drifting from the agreed spec. The harness wraps every build step between two deterministic gates and lets an adversarial reviewer attack it.
Missing values, extreme inputs, shape mismatches and leakage paths. Each finding is reproduced locally, then locked in as a test.
Only an explicit PASS from the executable gates admits the next stage — never the coding agent's own report.
The sandwich now runs unattended across all our projects: one step per AI session, two separate audits, automatic retries.
FAIL → a new session with raised effort reads the failure rack and patches; 4th failure stops for a human decision.
Chosen per product; most tools run fully on the user's PC with no file uploads.
Claude Code (hooks, MCP, sub-agents), OpenAI & Gemini APIs, PowerShell runner, Python tooling, memory & CPU guards
Python, TypeScript, Kotlin, Rust, C#, PowerShell
Tauri 2 + React, PySide6 / PyQt6, pywebview, PyInstaller, signed release ZIPs with SHA-256
PyTorch + CUDA, faster-whisper, TTS, Shap-E, TripoSG, FFmpeg (NVENC), Blender 5 add-ons, Remotion
FastAPI, Node.js, Vite, Next.js, Cloudflare Workers + D1, SQLite (WAL, FTS5)
three.js, Rapier, cannon-es, Unity 6 (C#), Android / Android Automotive OS
All numbers come from each project's own run logs; nothing without a record is claimed.
Diagnoses clean-room air-handling units from PLC history and SOPs; read-only, never writes to the PLC. FastAPI, SQLite, OpenAI.
Android Automotive concierge with a pure-Kotlin safety core that blocks interaction while driving. Demo build.
Identifying which language model wrote a text. Every experiment fixes its acceptance rule before seeing results.
Character TTS and narration to auto-subtitled, synced, mastered video.
Korean video to SRT drafts on the user's PC, with a crash-safe batch queue.
Audio + script to timed SRT using silence analysis — no AI API.
Screen recording and capture for Windows, with an automated lecture-recording mode.
One hotkey shows reply snippets beside the cursor and pastes into the original field.
Margin calculator, listing via API, order sheets for a live cross-border store; NStorePriceGuard watches competitor prices (changes locked).
Free shopping calculators today; next, a price-history deal service that filters fake discounts.
Text or image to PBR-textured GLB, imported straight into Blender.
Game asset pipeline with cost, license and SHA-256 recorded per asset.
Tray app watching PCs, servers, pipelines and MCP plugs; exposes a read-only MCP server.
Claude and Codex debate a decision in a CLI meeting with voice.
Offline coding assistant on a 12 GB GPU.
Web app that measures taste sensitivity against standard products and compares friends.
Calm multiplayer stone-stacking game.
Casual unboxing game; release build (AAB) ready, Google Play setup in progress.
Fixed 60 Hz physics decoupled from display rate, capped catch-up steps.
Korean trot sung by our original virtual singer — lip-sync, rendering and upscaling. Watch
Test counts are as recorded in each project's logs at the time noted there; some predate later changes.
Everything above was built and run on one desktop: RTX 3060 (12 GB), i5-12400F, 32 GB RAM. Code is fine; graphics are not. A lip-sync model alone needs ~11 GB, so heavy jobs queue one at a time through our own memory guard. With cloud GPUs and credits we would render full-length virtual-singer videos, batch seller videos, and generate 3D assets at full quality.