On the last post, I mentioned usuage of multiple github models, but made some more modifications, and now 2 new providers are also added to the boardroom.
STAGE 1 ARCHITECT CONFERENCE
Architect A (Mixtral/Groq) + Architect B (Llama/Groq) propose plans in parallel
Judge (GPT-4o) synthesises or proposes a more unpredictable plan
STAGE 2 IMPLEMENT
Engineer (GPT-4o) writes each file with full sibling context
STAGE 3 QUALITY LOOP (up to 8 rounds)
Reviewer A (Mixtral/Groq) + Reviewer B (Gemini) in parallel
Fixer applies merged feedback + Playwright interaction test
STAGE 4 QA REVIEW
Playwright clicks every button, tests every slider
QA Tester (GPT-4o) issues verdict: shippable / partially_usable / non_functional
STAGE 5 PUBLISH
New public GitHub repo created via API
GitHub Pages enabled -> live URL in under 60 minutes from cold start
STAGE 6 SELF-IMPROVE
CTO (Gemini) reads last 30 failure logs, proposes a surgical source patch
Validates Python syntax, commits - next build runs improved code