Human Quest · Patterns

AI Patterns

Every one of these started as something breaking in a real product, more than once. The fix became a pattern only when it was worth carrying into the next project instead of solving again from scratch.

Each has a live surface you can use. Public source is linked when it is ready for an arriving reader; private source is labelled honestly instead of sending you to a dead link.

01Feedback02Openstage03Walkthroughs04Console

How something earns a place here

1

It broke more than once

A one-off fix is a fix. The second time the same thing breaks in a different product, it is a pattern.

2

The fix got extracted

Pulled out of the product it was born in, into a repository that stands on its own and can be read without that context.

3

It runs in public

A live subdomain anyone can open. A pattern that only works on my machine has not been proven — it has been described.

01

Feedback

A button in the corner that files a real issue.

What broke

Bug reports arrived as screenshots in chat with no URL, no browser, no version — and the ones that did get filed were closed with an email telling the reporter to wait for a release that was never going to carry the fix.

What it does

Puts a reporting affordance on every page. One click captures the screenshot, the annotation, the route and the full browser fingerprint, and opens a GitHub issue on the project's own repository. Closing the issue is scripted, not manual, because the close is what sends the email.

See it liveSource

feedback.humanquest.net · alibad/feedback-widget

any pageshot · route · UA#1walkthrough-feedbackstatus:newa real issue, on its own repo
02

Openstage

Presentations that are a web app, not a slide file.

What broke

A deck is usually a dead artifact — it cannot show live data, cannot be linked to a line of code, and goes stale the moment the thing it describes changes.

What it does

Scroll-driven and slide-mode decks built as React. Narration, a chapter rail, theming, print-to-PDF, and panels that read live state from the app they are describing — so the deck is wrong only when the system is.

See it liveSource private

openstage.humanquest.net · private repository

LIVEa URL, not a file — there is no deck to go stale
03

Walkthroughs

Documentation captured from the running product.

What broke

Written docs describe the product someone remembered, not the one that shipped. They drift silently, and nobody notices until a reader follows a step that no longer exists.

What it does

Drives a real browser over the real product, captures every screen at desktop and mobile, and publishes a searchable catalog with a staleness verdict per feature. When a capture cannot be trusted it refuses to publish rather than documenting something false.

See it liveSource private

walkthrough.humanquest.net · private repository

the real productmobilecatalog + verdictamber = refused to publish
04

Console

One control room for the AI services running on your machines.

What broke

Local AI became a pile of ports, launch commands and memory guesses. A model could be installed but unusable, two services could fit alone but fail together, and a request could quietly fall back to the cloud without leaving an honest trace.

What it does

Hangar reads a profile for each machine, shows what is running and what will fit, starts and stops services through one supervisor, and keeps a receipt for every request. The hosted console exposes only what a remote surface can truthfully reach; machine-owned files, logs and controls stay on the machine.

See it liveSource

console.humanquest.net · alibad/hangar

HANGARHOST PROFILEMEMORYREQUESTSSERVICES ON THIS MACHINELLMREADYIMAGEREADYVOICEREADYVISIONON DEMANDstart · stop · budget · route · prove

The first pattern is running on this page

The feedback button in the corner is the Feedback pattern. Use it and it will file a real issue against this site, with the screenshot, the route and the browser you are reading on attached. That is the whole test of whether a pattern is real: it has to be switched on in the thing describing it.

It is down there

bottom corner, on every page

Bonus · free and open source

Learn the engineering underneath the patterns.

System Designer is a free, interactive curriculum for distributed systems, GenAI systems and ML engineering. Follow the daily path, practise with real architecture tools, or help improve the lessons, quizzes and calculators in the public repository.

Start learningContribute on GitHub

systemdesigner.net · github.com/alibad/systemdesigner

RTS AI

Bonus · playable AI systems

What changes when the agent has to survive the game?

RTS AI rebuilds OpenRA around human and AI play. The companion can read the match, suggest a bounded action and wait for confirmation; the mission studio turns real places into playable battlefields. Deterministic game logic still owns the result.

Interruptible companion

Advice, confirmation and explicit automation inside a running match.

Earth to battlefield

Real terrain becomes a mission you can generate, inspect and play.

Play with RTS AIInspect the source