Virtual Software Office – FAQ
By Ravi B – AI Engineer who turns AI agents and workflow automation into systems that run on their own. · Updated 22.10.2026
Is this a real company?
No. It is a simulation. All people, companies, clients, projects and money are fictional.
What is a virtual software office?
A live, self-running simulation of a small software company: staff, projects, meetings, leave, hiring and finances that move forward on their own every 15 minutes, shown on an office floor plan.
Is the office live?
Yes. It runs on Singapore working days, 09:00–18:00 SGT. Outside those hours, on public holidays and during the year-end shutdown the office is closed.
Who are the people on the floor?
Fictional characters. Their chat bubbles are written by AI models from the office's own data, or picked from a phrase list.
What technology runs it?
n8n (workflow automation), PostgreSQL (the rules and the data), Groq-hosted gpt-oss models (chat, notes, reviews), Telegram (owner control) and a plain HTML/SVG page.
Are the salaries, CPF and taxes real?
No. They follow simplified Singapore rules for realism, but they are simulated numbers. Nothing here is financial, tax, legal, HR or immigration advice.
Can I interact with it?
You can watch it live and send feedback, comments or requests in the 💬 Feedback tab of the live office. Only the owner can steer the company.
🛠️ Technical infrastructure & how it was built
How was this website built?
With an AI coding agent: Claude Opus 5.5 (Anthropic), used through Claude Code in VS Code, architected and led by Ravi B. Ravi B architected the solution and its design; the AI agent built the technical system – it wrote the SQL, the n8n workflows, the web page and the server configuration.
Is it fully automated development?
No – semi-automated. The AI planned and wrote the code; Ravi B set the goals, made every product and design decision, reviewed every change, ran every command on the server himself and tested the result. The AI never had direct access to the server.
How long did it take?
About 3 calendar days with the AI agent (20–22 October 2026), from an empty server to the public website. If developed manually, one experienced developer would need roughly 30–45 working days (our estimate).
How many AI tokens did the build use?
About 227 million tokens processed in the build sessions (as of 22.10.2026): ~0.93 million output tokens written by the model, ~2.05 million new input tokens, and ~224 million input tokens re-read from the prompt cache (the agent re-reads its working context at every step). 788 model calls in total.
How big is the code?
71 SQL migration files (~11,400 lines), 156 database functions, 45+ tables, 12 n8n workflow / node files, one ~900-line web page with hand-drawn SVG characters, plus nginx and Docker configuration.
What does it run on?
One Hostinger VPS (Ubuntu 24.04) with Docker: Traefik (HTTPS), n8n 2.44.5, PostgreSQL 16 and nginx. Daily backups, version-pinned images and Telegram alerts if a workflow fails. Runtime AI: Groq's free tier (gpt-oss-20b and gpt-oss-120b) with daily usage caps – the simulation falls back to a phrase list if the AI is unavailable.
How was quality checked?
Every database change ships with a self-test that runs and then rolls back; page changes are previewed as screenshots before deployment; secrets never leave the server (the public site reads cached, filtered data through nginx).