All projects
Q2 2026Apr 2026
LLM Consensus
Multi-Model AI Consensus Platform
Overview
LLM Consensus runs a 3-stage pipeline: up to 5 LLMs answer a query in parallel, anonymously rank peer responses, and synthesize a single final response. It streams stage-wise results in real time over Server-Sent Events, includes Supabase auth with conversation history and CRUD, Redis rate limiting, markdown/code/diagram rendering, voice input, and temporary chat sessions.
Highlights
- 3-stage LLM 'council': parallel answers, anonymous peer review, then chairman synthesis
- Real-time stage-wise streaming over Server-Sent Events
- Pick 2–5 models from any OpenRouter provider with Fast / Reasoning presets
- Supabase auth + conversation history, Upstash rate limiting, voice input
- Rich rendering: syntax-highlighted code, LaTeX math, and Mermaid diagrams
Tech stack
- Next.js
- FastAPI
- TypeScript
- Supabase
- Drizzle ORM
- Redis