Meme engine. AI engineering lab.
King Kongra makes up quotes nobody said, headlines nobody printed, and reviews nobody should trust. Underneath the jokes is a serious question: how do you tell whether a machine is actually being funny? This page is about both halves.
PROBABLY SAID(TM)
King Kongra
STACK
Elixir, Phoenix, LiveView, Ash, Postgres, pgvector, Oban, R2.
AI LOOP
Generate, rate, measure, compare, review, and promote style versions.
BOUNDARY
Elixir owns production. Python is reserved for offline analysis.
GOAL
Make humor quality observable without pretending automation replaces taste.
WHAT IT IS
A joke machine with receipts.
Using it is simple: type an absurd scenario, pick a style, get something back. Some memes come back as words, some as pictures. Everything it produces is fiction — the quotes are invented, the headlines never ran, and the people it sounds like never said any of it.
The harder part is underneath. Nothing about a language model tells you whether what it wrote is funny, so King Kongra measures instead of guessing: it records which memes people react to, locks each meme to the exact instructions that produced it, and can tell you afterwards whether a change to those instructions made things better or worse.
PRODUCT
Absurdist generation
Each style is written separately. A fake quote and a tabloid headline are different jokes, so they do not come from the same instructions.
QUALITY
Evaluation workbench
Reactions, skips, and format checks are turned into evidence about whether a style is landing — not a single vote, but enough of them to mean something.
JUDGMENT
Human review gates
Analysis can suggest a change to how a style writes. A person still has to agree before it goes live. Taste does not get automated here.
WHY IT EXISTS
A joke machine, built properly.
Most of what gets called AI engineering happens where nobody can see it. King Kongra is an attempt to build that part in the open, on something small enough to be honest about and silly enough to be worth using.
The site runs on Elixir and Phoenix, which are good at staying up and telling you what they are doing. Bigger AI frameworks are held back until there is evidence they would earn their complexity — knowing when not to add machinery is part of the exercise.
Built by MetaBureau and Stewart Milne as a practical AI product, a meme machine, and a place to make the invisible parts of AI engineering visible.