How Phil Talks to AI

A plain way in. Short phrases, and a model that learns them.

This is the piece beside Compass. It is about a person talking, and about the answer that comes back.

The main work

Michelangelo

Pre-execution AI governance. The control point before the output leaves the system.

Regulated AI needs more than logs after the fact. Michelangelo enforces admissibility at runtime — before an output can be used — and seals machine-readable evidence packs that can be replayed. Mapped to FDA 21 CFR Part 11 and EU GMP Annex 11. Built for high-consequence workflows including cleanroom environmental monitoring, deviations, and batch-release support. Provisional patent 63/924,367.

Raquel

AI wire compression — fewer tokens in, same answer out.

Raquel compresses language-aware payloads before they hit the tokenizer. Not gzip. Content-aware encoding that cuts inbound token load so providers and customers keep margin without changing the end-user experience.

Encoder Beacon Site Pulse Pitch Investor tour All Raquel 907

TokenScope

Real-time AI cost and session monitoring across the providers you actually use.

You cannot govern what you cannot see. TokenScope watches spend and sessions provider by provider — starting with the top active providers — so finance and engineering share one picture.

Multi-provider dashboard (active providers first)
Session-level cost and duration
Daily and period rollups
Provider status: active / quiet / error

In development

A History of AI

Six chapters. From Markov chains to Tuesday’s real threats.

A series for readers who want to understand how we got here — not just react to headlines.

Read the series Full portfolio

Tranquility

A shelved brief, back on the desk: thorium reactors to power training compute on the Moon. The hard stops were reactors, lift, time, and cost. Those stops have moved.

Read the papers and ask them questions

Now

Prove a Theorem That Does Not Exist

Four AIs, one trick prompt, and a governance lesson: you still sign what you publish.

Writing

All portfolio pieces →

Pick a title. That is the point.

Sycophancy as a Cost Variable

When flattery burns tokens — and the bill spikes 67%.

I Spent Yesterday Trying to Break AI

Graham’s number, odd/even traps, and where models start bluffing.

The Vending Machine That Couldn’t Say No

A live betta fish, and the scariest sentence of the year: he was not wrong.

And some fiction, too

Not everything here is a white paper. If you want a short story:

The Useful Mission — three characters, a boat, Frying Pan Shoals, 1962.

About

Pink House Technology builds practical control points for AI at work — governance before the output leaves the system, compression before tokens burn the budget, and visibility across the providers you actually use.

Phil Cheevers founded Pink House to close the gap between “we use AI” and “we can defend, measure, and afford how we use AI.” Michelangelo is the lead work. Compass, on this page, is a separate social contribution for vision-impaired people. The supply-chain book is in the portfolio.

Contact

Best for: partnerships, regulated QA, compression pilots, TokenScope early access

philcheevers@pinkhouse.com