<div align="center">

🤖 SACHI

Supervisor Agent for CrewAI Hierarchical Intelligence

A Hierarchical Multi-Agent AI System powered by CrewAI, Ollama and Streamlit

<p align="center"> <img src="assets/01-sachi-overview.png" width="100%"> </p>

</div>

---

📖 Introduction

SACHI (Supervisor Agent for CrewAI Hierarchical Intelligence) is a hierarchical multi-agent AI system built using CrewAI, Ollama, and Streamlit.

Unlike traditional AI assistants where a single LLM attempts every task, SACHI introduces a Supervisor Agent responsible for intelligently understanding the user's request, selecting the most appropriate specialist agent, delegating execution, and finally returning a synthesized response.

The system demonstrates how hierarchical task delegation can make AI systems more scalable, modular, and easier to maintain.

This project was developed during the

Summer Training Programme on Machine Learning & Agentic AI

Electronics & ICT Academy, IIT Roorkee

---

🎯 Why SACHI?

Large Language Models are powerful, but expecting one model to solve every problem is inefficient.

Real organizations solve problems differently.

A manager receives a task...

↓

Understands the requirement...

↓

Assigns it to the right specialist...

↓

Reviews the output...

↓

Delivers the final result.

SACHI follows exactly the same philosophy.

Instead of asking one AI to know everything,

a Supervisor Agent dynamically delegates work to specialized AI agents depending on the user's request.

---

✨ Key Features

| Feature | Description | |---------|-------------| | 🧑‍💼 Supervisor Agent | Analyzes every prompt and decides which agent should execute it | | 🧠 Data Scientist Agent | Handles ML, AI, Python and statistics related queries | | 📊 Data Analyst Agent | Handles reporting, dashboards and business analytics | | 🔀 Hierarchical Delegation | Uses CrewAI's Process.hierarchical architecture | | 💬 Streamlit Chat UI | Interactive conversational interface | | 🖥 Local LLM | Runs completely on Ollama using llama3.2 | | ⚙ YAML Configuration | Agents and Tasks configured through YAML | | 🚀 Extensible Design | Easily add new specialist agents |

---

🏗 Architecture

flowchart TD

User([User Prompt])

Supervisor[Supervisor Agent]

DS[Data Scientist Agent]

DA[Data Analyst Agent]

Response([Final Response])

User --> Supervisor

Supervisor --> DS

Supervisor --> DA

DS --> Response

DA --> Response

---

⚙ Tech Stack

| Layer | Technology | |---------|------------| | Multi-Agent Framework | CrewAI | | LLM Runtime | Ollama | | Model | llama3.2 | | Frontend | Streamlit | | Configuration | YAML | | Language | Python 3.10+ | | HTTP | Requests |

---

📂 Project Structure

SACHI
│
├── app.py
├── config
│   ├── agents.yaml
│   └── tasks.yaml
├── assets
├── requirements.txt
├── README.md
└── LICENSE

📦 Requirements

• Python 3.10+
• Ollama Installed
• llama3.2 Model
• Git
• Internet connection (only for installing dependencies)

---

🚀 Installation

1️⃣ Clone the Repository

git clone https://github.com/Prannoybuilds/sachi.git
cd sachi

---

2️⃣ Create Virtual Environment

Windows

python -m venv venv

venv\Scripts\activate

Linux / macOS

python3 -m venv venv

source venv/bin/activate

---

3️⃣ Install Dependencies

pip install -r requirements.txt

---

4️⃣ Install Ollama

Download from

https://ollama.com/download

After installation

ollama pull llama3.2

ollama serve

---

▶ Running the Application

streamlit run app.py

Open your browser

http://localhost:8501

Once connected,

the sidebar should display

✅ Ollama Connected

---

🔄 Workflow

User Prompt
      │
      ▼
Supervisor Agent
      │
      ▼
Prompt Analysis
      │
      ▼
Task Delegation
      │
      ├──────────────► Data Scientist Agent
      │
      └──────────────► Data Analyst Agent
               │
               ▼
     Specialist Response
               │
               ▼
 Supervisor Synthesizes Result
               │
               ▼
      Final Response

---

📸 Project Demonstration

Complete Execution

The complete execution flow of SACHI from user prompt to delegated response.

<p align="center"> <img src="assets/01-sachi-overview.png" width="100%"> </p>

---

🏠 Home Screen

The Streamlit interface where users interact with the Supervisor Agent.

<p align="center"> <img src="assets/02-home-screen.png" width="95%"> </p>

---

📊 Supervisor Delegates Task

The Supervisor analyzes the incoming prompt and delegates it to the appropriate specialist agent.

<p align="center"> <img src="assets/03-data-analyst-delegation.png" width="95%"> </p>

---

🚀 Agent Execution

The delegated specialist begins processing the assigned task.

<p align="center"> <img src="assets/04-agent-started.png" width="95%"> </p>

---

🔀 Hierarchical Delegation Trace

CrewAI verbose logs showing the Supervisor's reasoning and delegation process.

<p align="center"> <img src="assets/05-delegation-trace.png" width="95%"> </p>

---

✅ Final Crew Response

Final synthesized output generated after successful collaboration between Supervisor and Specialist Agents.

<p align="center"> <img src="assets/06-crew-completed.png" width="95%"> </p>

---

🌟 Key Highlights

• Intelligent Supervisor Agent

• Dynamic Task Delegation

• Hierarchical Multi-Agent Architecture

• Fully Local AI Execution

• Zero API Cost

• Modular YAML Configuration

• Interactive Streamlit Interface

• Extensible Agent Framework

• Transparent Execution Logs

• Open Source

• 📊 Demonstration Results

SACHI successfully demonstrates a production-style hierarchical multi-agent workflow capable of intelligently routing user requests to specialized AI agents.

Demonstrated Capabilities

• ✅ Intelligent prompt understanding
• ✅ Supervisor-driven task delegation
• ✅ Dynamic specialist selection
• ✅ Local LLM inference with Ollama
• ✅ Transparent execution logs
• ✅ Interactive Streamlit chat interface
• ✅ Modular YAML-based agent configuration
• ✅ End-to-end CrewAI orchestration

---

🎓 Learning Outcomes

Building SACHI provided hands-on experience in:

• Designing hierarchical multi-agent AI systems
• Understanding CrewAI's Process.hierarchical
• Building supervisor-specialist agent architectures
• Local LLM deployment using Ollama
• YAML-driven prompt engineering
• Streamlit application development
• Agent orchestration and execution tracing
• Managing modular AI workflows
• Building scalable agentic AI applications

---

🚀 Future Improvements

The architecture has been intentionally designed to be extensible.

Future enhancements include:

• 🤖 Additional specialist agents (Research, Coding, Finance, Legal)
• 🌐 Multi-LLM support (OpenAI, Claude, Gemini, DeepSeek)
• 📄 Document Question Answering
• 📚 RAG (Retrieval-Augmented Generation)
• 🧠 Long-term Memory using Vector Databases
• 🔊 Voice-based interaction
• 📂 Multi-file document analysis
• ⚡ Streaming responses
• 🐳 Docker deployment
• ☁ Cloud deployment on AWS/Azure/GCP
• 📊 Agent performance analytics dashboard

---

🎯 Applications

SACHI can serve as the foundation for:

• AI Assistants
• Enterprise Knowledge Systems
• Customer Support Agents
• Research Assistants
• Business Intelligence Platforms
• Internal Company Chatbots
• Educational AI Tutors
• Decision Support Systems
• Multi-Agent Enterprise Automation

---

💡 Why Hierarchical Agents?

Traditional AI systems rely on a single model for every task.

SACHI demonstrates a more scalable approach inspired by real-world organizations, where a Supervisor coordinates specialized experts to solve complex problems efficiently.

This architecture improves:

• Scalability
• Maintainability
• Specialization
• Transparency
• Extensibility
• Decision Quality

making it suitable for building next-generation Agentic AI systems.

---

🤝 Contributing

Contributions are welcome!

If you would like to improve SACHI, feel free to:

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Push the branch
5. Open a Pull Request

Suggestions, bug reports, and feature requests are always appreciated.

---

📄 License

This project is licensed under the MIT License.

See the LICENSE file for additional details.

---

👨‍💻 Author

Prannoy Sen

B.Tech Computer Science & Engineering

Manipal University Jaipur

Connect with me

• GitHub: https://github.com/Prannoybuilds
• LinkedIn: https://www.linkedin.com/in/prannoysen/

---

🙏 Acknowledgements

Special thanks to:

• Electronics & ICT Academy, IIT Roorkee
• CrewAI for providing an exceptional multi-agent framework.
• Ollama for enabling fully local LLM execution.
• Streamlit for rapid interactive application development.
• The open-source AI community for continuously advancing Agentic AI research.

---

<div align="center">

⭐ If you found this project useful, consider giving it a Star!

SACHI

Supervisor Agent for CrewAI Hierarchical Intelligence

Building the Future of Hierarchical Agentic AI 🚀

</div> # 🛣️ Roadmap

• [x] Hierarchical Multi-Agent Architecture
• [x] Local LLM Integration
• [x] Streamlit Interface
• [x] YAML Agent Configuration
• [ ] Retrieval-Augmented Generation (RAG)
• [ ] Long-Term Memory
• [ ] Multi-LLM Support
• [ ] Docker Deployment
• [ ] Kubernetes Deployment
• [ ] Web Search Agent
• [ ] Code Execution Agent
• [ ] Research Agent
• [ ] Voice Assistant

---

📸 Sample Prompts

Try asking SACHI:

🧠 Data Science

Explain the bias-variance tradeoff.

Write a Random Forest implementation in Python.

How does XGBoost differ from LightGBM?

---

📊 Data Analysis

Create a sales dashboard for monthly revenue.

Analyze customer churn trends.

Generate KPIs for an e-commerce business.

---

🤖 AI & Machine Learning

Explain transformer architecture.

Difference between CNN and Vision Transformer.

Build an ML pipeline for fraud detection.

---

⭐ Project Status

🚀 Actively maintained and continuously improved as part of my Agentic AI learning journey.

---

❤️ Support

If you found this repository useful:

⭐ Star the repository

🍴 Fork the repository

🛠️ Contribute improvements

📢 Share it with others

Every contribution helps improve SACHI.