Qurultai is a sovereign, multi-agent decision support platform designed to modernize public administration. Named after the ancient Turkic council of leaders, Qurultai simulates a "Digital Cabinet" where AI agents represent specific government ministries.
These agents analyze policy documents, detect conflicts between legal frameworks, and collaborate to generate consensus-based recommendations. It transforms the opaque, siloed process of government decision-making into a transparent, auditable, and efficient digital workflow.
Government decision-making is plagued by Institutional Fragmentation:
- Siloed Data: Ministries operate independently; decisions by one often violate the regulations of another.
- Opacity: Citizens and businesses often cannot understand why a decision was made or which laws apply.
- Latency: Inter-ministerial coordination is manual, paper-based, and slow.
Qurultai introduces an Agentic Orchestration Layer over government data.
- Input: A user (Citizen, Business, or Official) submits a query or document.
- Council: Specialized AI agents (Finance, Justice, Ecology) analyze the input against their specific legal knowledge bases.
- Debate: Agents identify conflicts and propose solutions in a transparent "Group Chat" format.
- Consensus: A Synthesizer agent compiles the debate into a structured, legally referenced PDF report.
- The Council Chamber: A unique chat interface where users observe AI agents debating and collaborating in real-time.
- Agent Orchestration Panel: An admin dashboard to create, configure, and deploy specialized agents (e.g., assign "Tax Code" to the Finance Agent).
- Sovereign RAG Engine: Retrieval-Augmented Generation runs locally using Ollama, ensuring sensitive government data never leaves the infrastructure.
- Explainable AI: Every decision is traced back to specific articles in the law. No "Black Box" answers.
- PDF Report Generation: Automatic creation of official-style documents for government circulation.
This project uses a modular, sovereignty-first architecture:
| Component | Technology |
|---|---|
| Orchestration | LangGraph (State Machine for Cyclic Debates) |
| Frontend | Vue.js 3 + Vite + PrimeVue |
| Backend | FastAPI (Python) |
| Local LLM | Ollama (Llama 3 / Mistral) |
| Vector Store | ChromaDB |
| Database | SQLite |
| Embeddings | Nomic Embed Text (via Ollama) |
graph LR
A[Vue.js Client] --> B[FastAPI Gateway]
B --> C{LangGraph Router}
C --> D[Agent: Finance]
C --> E[Agent: Ecology]
C --> F[Agent: Justice]
D --> G[ChromaDB RAG]
E --> G
F --> G
D <--> H[Debate Loop]
E <--> H
F <--> H
H --> I[Synthesizer Agent]
I --> J[Final Report PDF]