← All projects
AI / ML2026 / LLM Council

LLM Council

A local multi-LLM orchestration system. Requests are concurrently processed by multiple LLMs, which then critique and rank each other's outputs, with a final synthesis generated by a chairman LLM.

My role

Own development. Implementation details below.

Scope

LLM Council

Links

No public link published

01 / The challenge

What the product does

An orchestration sandbox engineered to produce high-accuracy outputs for complex prompts. An inquiry is routed to 3-5 different LLMs (Claude, GPT, Gemini) concurrently. The agents then critique and score each other's answers, after which a final synthesizer agent generates a master response.

02 / The solution

LLM Council

An orchestration sandbox engineered to produce high-accuracy outputs for complex prompts. An inquiry is routed to 3-5 different LLMs (Claude, GPT, Gemini) concurrently. The agents then critique and score each other's answers, after which a final synthesizer agent generates a master response.

  • Cross-model review
  • Chairman LLM
  • Runs locally
Technology
Implementation details
  • Gracefully handling timeouts or rate limits of individual model providers to guarantee final output compilation.
  • Reducing API expenditures by stripping redundant context windows from grading iterations.
Next projectTropicVibe

Your next good product starts here

Have a project?
Let’s figure it out.

Bring an idea, a prototype or a process you have wanted to simplify.

Message on Telegram Build a brief

A short brief helps frame the conversation.
No commitment to start a project.