AI / Agent2026
LLMGroupChat
A consensus engine where multiple LLMs debate and vote to pick stocks.
Stack
- Dart
- Supabase
- Deno
- LLM APIs
The problem
A single LLM's stock pick is a coin-flip. I wanted to see whether a *panel* of models — each proposing and scoring ideas — could produce a more reliable signal than any one of them alone.
What I built
- A multi-model consensus pipeline: chats → cycles → iterations → propositions.
- MOVDA scoring and a 2-in-a-row winner rule to decide when the panel has actually converged.
- Supabase edge functions (Deno 2 runtime) orchestrating the model calls and persistence.