PRODUCT
Absurdist generation
Fake quotes, headlines, reviews, manifestos, and image prompts are handled as style-specific outputs rather than one generic prompt.
ABOUT KING KONGRA
Meme engine. AI engineering lab.
King Kongra is a production Phoenix/Ash meme generator that doubles as an AI engineering implementation exploration project. The fun surface creates fake quotes, tabloid headlines, unhinged reviews, mini manifestos, and image-backed meme cards. The deeper work is the machinery around AI quality: safety gates, style versioning, semantic caching, feedback capture, evaluation, and offline analysis.
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.
The consumer product is deliberately simple: type an absurd scenario, choose a style, and get a meme-ready result. Some outputs are text-first. Some are image-backed. All of them are designed around fictional, archetypal, or clearly reviewed attribution modes so the joke stays legible and safe.
The engineering project underneath is more ambitious. King Kongra treats comedy generation as an AI quality problem. It captures user ratings, pins generations to immutable style versions, tracks cache behavior, and builds historical evaluation reports so style changes can be judged by evidence instead of vibes alone.
QUALITY
Evaluation workbench
Ratings, skip rates, weighted signals, format checks, effective sample size, and dominance warnings turn messy feedback into usable evidence.
JUDGMENT
Human review gates
Candidate prompt improvements can be proposed by analysis, but promotion stays manual through immutable style versions and review.
WHY IT EXISTS
A portfolio project that earns its architecture.
King Kongra is built to be interviewable without becoming architecture theater. The production path stays in Phoenix and Ash because the BEAM is a strong fit for reliability, request handling, telemetry, and operator workflows. Python enters only where it is strongest: offline dataset analysis, report generation, and exploratory AI evaluation.
DSPy and heavier AI frameworks are deliberately deferred until the data proves they are worth the complexity. That restraint is part of the point: the project is about knowing when to add machinery, not collecting frameworks like novelty mugs.
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.