BACK TO MANUAL / PROJECTSMVM LABS · CASE STUDY 02
PROJECT DOSSIER // Multi-model deliberation app

LLM Council

TECHNICAL OVERVIEW

LLM Council is a local-first web app that sends the same prompt to multiple Ollama models, lets them critique and rank each other, then synthesizes a final answer through a chairman model.

ARCHITECTURE DETAILS

A multi-stage AI deliberation system where several models respond independently, review anonymized peer answers, vote on quality, and pass their evaluations to a chairman model for final synthesis. Supports both local Ollama and Ollama Cloud, stores conversations as JSON, and streams final answers over Server-Sent Events.

FIG 01 // SYSTEM DIAGRAM SCHEMATIC
USER PROMPTLocal web UIMODEL ALlama3MODEL BPhi3MODEL CMistralPEER REVIEWAnonymized votingCritique loopsCHAIRMANConsolidationFinal synthesisSSE STREAMStreamed response
PROJECT SPECS
CategoryAI SYSTEMS / LOCAL-FIRST AI
StatusACTIVE / EXP
UpdatedJUN 2026
LAUNCH CODE REPO
SYSTEM DESIGN NOTE

Shows systems thinking around model orchestration, peer evaluation, ranking, fallback behavior, local/cloud runtime switching, and AI workflow design.

TECHNOLOGY STACK
FastAPIReactViteOllamahttpxServer-Sent EventsJSON Storage
METADATA TAGS
#llms#ollama#multi-agentreasoning#fastapi#sse#local-firstai