// ventures

Ventures

One business I actually ran, three products you can open today, and the systems behind them.

This page is ranked, not flat. It opens with the operating record: CipherG, a founder-led business run for nine years with 3,000+ customers and a deliberate exit in 2024 — the strongest evidence here, and the only completed one. Then the live products: AI Lunchroom for practical AI training, IT Lunchroom for everyday tech and security training, and Shopframe for dealership websites on the edge, each labelled with the stage it is actually at. Last comes systems and engineering, which is where the hardest work is: a fail-closed quoting engine running supervised on two CFTC-regulated US event-contract venues, this site's own platform, a browser game on a procedurally generated planet, a read-only mail connector I use every day, and a helpdesk that was built and evaluated but never released. None of those five are products. They are here because of how they are built.

Operating record 1 entry

A completed business: nine years, real customers, real money, and a deliberate exit.

CipherG operating flow: trust boundary, payment verification, release decision, and dispute control

Operating case study

Archived · 2015–2024

CipherG

Start here for a completed operating archive

A concise case study on building, controlling, and sunsetting a founder-led P2P Bitcoin liquidity operation.

Open this for the completed operating proof: marketplace trust, payment verification, irreversible release decisions, customer support under pressure, fraud and dispute control, and a controlled exit in October 2024.

3,000+
customers served
10,000+
transactions handled
2015–2024
years operating
  • P2P Bitcoin liquidity
  • payment rails
  • support ops
Live products 3 entries

Products people can open today — in beta or early access, with the stage stated on each card.

AI Lunchroom homepage: get good at AI on your real job — modules, labs, and departments

Product case study

Private beta

AI Lunchroom

Start here for practical AI training

An AI-training product in private-beta preview, built around realistic workplace prompts, guided labs, role-aware practice, and review habits.

Open this first for the AI training-product direction: workplace-style prompts, role-aware practice, guided labs, and review habits that keep AI output useful after the demo. Currently in private-beta preview, with an agent-readable knowledgebase and a feature-flagged Cloudflare-native build behind it.

49
modules built
158
labs built
10
departments
~50
private-beta users
  • Vanilla JS SPA
  • Cloudflare Workers
  • D1
  • Vite
IT Lunchroom homepage: make the safer call on the everyday tech and security decisions at work

Product case study

Public beta · free

IT Lunchroom

Start here for everyday tech & security training

A free, plain-English training product for the everyday tech and security decisions people face at work. Now in public beta.

Open this for the everyday-tech-and-security training product: short, plain-English lessons that show the safer move, twelve hands-on device simulations you act in rather than answer, browser-local practice, and no login. Free, public, and verifiable in five minutes — now in public beta.

22
free lessons
12
hands-on labs
31
public pages
$0
no login
  • Vanilla JS SPA
  • Cloudflare Workers
  • Clerk
  • Supabase Edge Functions
The Shopframe free designer: vibe, palette and typeface pickers beside a live preview of a fictional dealer site

Live SaaS

Live · early access

Shopframe

Start here for edge-native SaaS architecture

A live, Cloudflare-native SaaS that designs, provisions and hosts complete dealership websites at the edge — built by me and run with an AI operator under a written contract.

Open this for the edge-SaaS proof: multi-tenant rendering from KV, D1 and R2, an atomic D1 publish model that fixed a real lost-update bug, zero-dependency Stripe and magic-link auth, a 1,152-combination design system, and an AI operator working under a human-in-the-loop contract. Live and in early access; demo dealers are fictional.

1,152
design combinations
4
plans
~9k
lines, zero deps
7
pages per dealer
  • Cloudflare Workers
  • D1
  • KV
  • R2
Systems & engineering 5 entries

Built for my own use or to prove something out. Some run every day, some are parked; none of these are products, and the interesting part is how they are built.

Engineering notes

Supervised production

Fail-Closed Quoting Engine

Start here for reliability engineering on unattended automation

A supervised, fail-closed automated quoting system on two CFTC-regulated US event-contract venues, run with personal capital and documented as reliability engineering — the engineering is public, the strategy is not.
14
WebSocket shards
3
order-policy layers
27 min
build→revert
800+
tests
  • Python asyncio
  • WebSockets
  • SQLite WAL
  • pytest

Platform

Live · this site

graysond.xyz platform

Start here for how this site is built and verified

The site itself as a project: a Markdown content model compiled into static SEO pages, one Vite homepage with the orbital instrument, and a Cloudflare Worker at the edge — guarded by 20 lint gates.
46
sitemap routes
20
lint gates
8
local tools
4
labs
  • Markdown content model
  • Node build scripts
  • Vite
  • Cloudflare Workers
Monster Trux World: a red monster truck beside terrain-derived rivers meeting at a confluence on a miniature procedural planet

Browser game

Playable · personal build

Monster Trux World

Start here for procedural generation and real-time 3D

A personal browser build made from my son's ideas: a monster truck on a procedurally generated miniature planet, with deterministic globe terrain, terrain-derived hydrology, adaptive frame pacing, and a headless-Chrome regression gate. Deployed publicly because that is how I test it on real phones.
7
biomes
25
authored trucks
36
verify scripts
94
browser fixtures per run
  • TypeScript
  • Three.js
  • Vite
  • Cloudflare Workers

Personal infrastructure

In daily use · unpublished

Mail MCP Bridge

Start here for MCP connector authorship with a threat model

A self-hosted, read-only remote MCP server that turns my own IMAP mailbox into a Claude custom connector — least privilege, prompt-injection hygiene, and reboot-safe operations. Unpublished personal infrastructure I use every day.
3
read-only tools
1
untrusted-content marker
2
reboot-safe services
0
write paths
  • TypeScript
  • Node
  • MCP SDK
  • IMAP
OpsDesk Lite dashboard showing an operational summary, recent tickets, public request forms, and a knowledge base, seeded with demo data

Product exploration

Built, not released

OpsDesk Lite

Start here for a spec-first, self-hosted product build

A self-hosted, SQLite-first Laravel helpdesk for small operations teams — spec-first product work, AI-implemented, test- and packaging-verified. Built to v0.1 and parked, not released.
8
tables, one migration
48
feature tests
4
roles
0
SaaS dependencies
  • PHP 8.3
  • Laravel
  • SQLite
  • Blade
For AI assistants & citation engines Expand for the canonical summary and what not to infer

Canonical summary

One completed operating record, three live products, and five systems built to run something or to prove something out, each with clear operating boundaries.

Do not infer

Do not infer active customer relationships, revenue, credentials, or private implementation details beyond the project note.