Three-model routing system
Quentin040507 (刘全) · Data Science · Intake 2026 · AI tools
Repository: Quentin040507/multi-llm-router
Multi-model collaborative LLM intelligent routing prototype: automatic routing, round table diagnosis, raised hand voting, one entrance with multiple models
- llm
- ai
- python
- fastapi
- deepseek
- qwen
- kimi
Project overview
Project Background and Positioning There are obvious shortcomings in the ability of a single large model: some are good at code and mathematical reasoning, but writing Chinese copywriting is rigid; some Chinese writing is fluent, but they are prone to errors in logic and calculations.In daily use, users often have to judge for themselves which model to ask this question, switch back and forth between different platforms, and cannot directly compare or synthesize the answers of multiple models.This is what this project is trying to solve: do a "multi-model collaboration" routing and orchestration prototype, use one portal to use the advantages of multiple models at the same time - Intelligent Routing Automatically determine the problem domain and distribute it to the most proficient model; Roundtable consultation Multiple models engage in real-time debates and mutual comments. The host integrates the viewpoints of the three parties to produce a complete answer. Vote by raising hands Objective questions adopt a majority vote mechanism to reduce the probability of errors in a single model. It is suitable for students and individual developers for daily Q&A, writing copy, and assisting in code writing. It is also suitable for those who want to compare the effects of large models horizontally or learn the idea of "multi-model orchestration/agent routing". ⚠️ This project is Experimental student prototypes to explore routing and orchestration ideas for multi-model collaboration, Not recommended for direct use in a production environment (No authentication and quota management, interface key documentation is stored locally, no high availability/monitoring). How to use Three models are integrated (DeepSeek is responsible for code and mathematical reasoning, Qwen for Chinese writing, and Kimi for general dialogue and long text understanding), all of which are OpenAI-compatible interfaces. It provides dual entry points for web Windows and CLI. Just fill in the API key in the upper right corner to use. Function Intelligent routing: automatically determine the problem area, choose the best model, fail to automatically demote Roundtable consultation: Three models real-time debate + moderator summary, the process can be reviewed Vote by show of hands: objective majority vote Multi-Session and History: Multi-Task Switching, Turn Off and Re-Open Witho…