🧠 Astraeus AI

Autonomous Multi-Agent Research & Evaluation Engine

Astraeus AI is a modular, production-oriented multi-agent research engine that performs structured research, verification, writing, critique, and quality evaluation using large language models.

Built as a personal AI systems engineering project to explore autonomous agent orchestration, iterative refinement loops, and research quality scoring.

---

🚀 Overview

Astraeus AI simulates a structured research workflow using multiple AI agents:

1. Research Analyst
2. Fact Verification Specialist
3. Technical Writer
4. AI Output Critic

The system:

• Generates structured research
• Verifies factual consistency
• Produces a formatted technical report
• Critiques and scores the output
• Iteratively refines until quality threshold is met

It exposes:

• 🖥 FastAPI backend (LLM orchestration)
• 🌐 Streamlit dashboard (interactive UI)
• 🔁 Iterative evaluation loop
• 📊 Score extraction and weakness detection

---

🏗 Architecture

Multi-Agent Pipeline

flowchart TD
    A([🧑 User Topic]) --> B

    B[🔍 Research Agent\nGathers relevant information\nfrom multiple sources]
    B --> C

    C[✅ Verification Agent\nFact-checks and validates\nresearch accuracy]
    C --> D

    D[✍️ Writer Agent\nSynthesizes research into\na structured article]
    D --> E

    E[🎯 Critic Agent\nEvaluates quality, clarity,\nand depth of the output]
    E --> F

    F{📊 Score Extraction\n+ Weakness Analysis}

    F -- Score ≥ Threshold --> G([✅ Final Output])
    F -- Score < Threshold --> H[🔄 Iterative Refinement]
    H --> D

    style A fill:#6366f1,color:#fff,stroke:none
    style G fill:#22c55e,color:#fff,stroke:none
    style F fill:#f59e0b,color:#fff,stroke:none
    style H fill:#ef4444,color:#fff,stroke:none
    style B fill:#1e293b,color:#94a3b8,stroke:#334155
    style C fill:#1e293b,color:#94a3b8,stroke:#334155
    style D fill:#1e293b,color:#94a3b8,stroke:#334155
    style E fill:#1e293b,color:#94a3b8,stroke:#334155

Loading

Backend

• FastAPI REST API
• CrewAI multi-agent orchestration
• Local LLM via Ollama
• Score parsing & weakness detection modules

Frontend

• Streamlit dashboard
• Glassmorphism UI
• Sectioned report visualization
• Export (Markdown / JSON)

---

🧰 Tech Stack

• Python 3.10+
• CrewAI
• Ollama (Local LLMs)
• FastAPI
• Uvicorn
• Streamlit
• Pydantic

---

📂 Project Structure

User Topic
    ↓
Research Agent        →  Gathers information from multiple sources
    ↓
Verification Agent    →  Fact-checks and validates accuracy
    ↓
Writer Agent          →  Synthesizes into a structured report
    ↓
Critic Agent          →  Evaluates quality, clarity, and depth
    ↓
Score Extraction + Weakness Analysis
    ↓
Iterative Refinement  →  Loops back to Writer if score < threshold
    ↓
Final Output

---

⚙️ Installation

1️⃣ Clone the repository

git clone https://github.com/YOUR_USERNAME/astraeus-ai.git
cd astraeus-ai

2️⃣ Create virtual environment

python -m venv .venv source .venv/bin/activate # Linux / Mac .venv\Scripts\activate # Windows

3️⃣ Install dependencies

pip install -r requirements.txt

🧠 LLM Setup (Ollama)

Install Ollama and pull a model: ollama pull llama3.1:8b ollama pull llama3.1:8b

▶️ Running the Backend

From project root: uvicorn app.api:app --reload

API docs: http://127.0.0.1:8000/docs

🌐 Running the Dashboard

In a new terminal: streamlit run dashboard/app.py

Access: Access: http://localhost:8501

📊 Example Output Format

SCORE: 8

STRENGTHS:

• ...
• ...

WEAKNESSES:

• ...
• ...

IMPROVEMENTS:

• ...
• ...

🎯 Key Features

• Modular multi-agent architecture
• Iterative quality refinement loop
• Automatic score extraction
• Weakness counting engine
• Research critique system
• Exportable reports
• Production-style backend/frontend separation

🧪 Design Goals

• Explore autonomous AI agent collaboration
• Simulate structured research workflows
• Implement measurable output quality evaluation
• Build system-level AI engineering skills

👨‍💻 Future Improvements

• RAG integration
• Vector database memory
• Multi-model ensemble evaluation
• Advanced metric scoring (precision/recall scoring engine)
• Deployment to cloud infrastructure
• CI/CD integration

👨‍💻 Author

Devdutt S

Personal upskilling project focused on building real-world AI systems architecture.

📜 License

MIT License