🤖 CrewAI Analytics Assistant using Ollama & MCP

A Multi-Agent AI Analytics Assistant built using CrewAI, Ollama, Model Context Protocol (MCP), and Streamlit.

The application leverages a hierarchical multi-agent architecture where a Supervisor Agent intelligently delegates user requests to specialized Data Analyst and Data Scientist agents. It provides conversational analytics capabilities including dataset profiling, SQL validation, KPI recommendations, dashboard suggestions, and machine learning planning using locally hosted LLMs.

---

🚀 Features

• Multi-Agent AI system using CrewAI
• Hierarchical agent delegation
• Local LLM inference using Ollama
• Interactive Streamlit chat interface
• Model Context Protocol (MCP) integration
• CSV dataset profiling
• SQL query validation
• Data quality assessment
• KPI recommendation
• Dashboard layout suggestion
• Machine learning problem recommendation
• Feature engineering suggestions
• ML evaluation metric recommendation
• End-to-end ML pipeline planning

---

🏗️ System Architecture

                     User
                       │
                       ▼
              Streamlit Interface
                       │
                       ▼
              Supervisor Agent
                       │
        ┌──────────────┴──────────────┐
        ▼                             ▼
 Data Analyst Agent          Data Scientist Agent
        │                             │
        └──────────────┬──────────────┘
                       ▼
              Function Tools Layer
                       │
                       ▼
                 MCP Server
                       │
     ┌───────────┬────────────┬─────────────┐
     ▼           ▼            ▼             ▼
 CSV Tools   SQL Tools   KPI Tools    ML Tools

---

📂 Project Structure

crewai-analytics-assistant/
│
├── agents/
│
├── config/
│
├── function_tools/
│
├── mcp_server/
│   ├── sample_data/
│   └── tools/
│
├── screenshots/
│   ├── home.png
│   ├── delegation.png
│   └── final_response.png
│
├── tests/
│   ├── test_analyst.py
│   ├── test_csv_profile.py
│   ├── test_data_quality.py
│   ├── test_crew_tools.py
│   ├── test_scientist.py
│   ├── test_sql.py
│   └── test_supervisor.py
│
├── app.py
├── requirements.txt
├── README.md
└── .gitignore

---

🛠️ Tech Stack

Languages

• Python

AI Frameworks

• CrewAI
• Ollama

Protocol

• Model Context Protocol (MCP)

Data & Analytics

• Pandas
• DuckDB
• SQL

UI

• Streamlit

Configuration

• YAML

---

👥 Agent Responsibilities

🧠 Supervisor Agent

Responsible for:

• Understanding user intent
• Delegating tasks to specialist agents
• Coordinating execution
• Validating outputs
• Returning the final response

---

📊 Data Analyst Agent

Responsible for:

• CSV Profiling
• Data Quality Analysis
• SQL Validation
• KPI Recommendation
• Dashboard Suggestions
• Business Insights

---

🤖 Data Scientist Agent

Responsible for:

• Machine Learning Recommendation
• Feature Engineering
• Evaluation Metrics
• ML Pipeline Planning
• Predictive Analytics

---

🔧 MCP Server Tools

| Tool | Purpose | |------|---------| | Profile CSV | Analyze dataset structure | | Validate SQL | Validate SQL queries | | Run DuckDB Query | Execute SQL queries on CSV | | Detect Data Quality | Missing values, duplicates, cardinality | | Generate KPI Catalog | Business KPI recommendations | | Recommend ML Use Cases | ML recommendations | | Generate Report | Markdown report generation |

---

📸 Application Preview

🏠 Home Screen

Configure the local LLM and interact with the analytics assistant.

---

🔄 Hierarchical Agent Delegation

Supervisor agent delegates work to specialist agents while tracking execution.

---

📊 Final Analytics Report

Dataset profiling, KPI recommendations, dashboard suggestions, and business insights generated from the uploaded CSV dataset.

---

💬 Example Prompts

CSV Analysis

Analyze the CSV dataset located at

mcp_server/sample_data/customers_sample.csv

Profile the dataset, identify any data quality issues,
recommend suitable KPIs, and suggest a dashboard layout.

---

SQL Validation

Validate this SQL query:

SELECT * FROM dataset LIMIT 10

---

Machine Learning Planning

The business wants to predict customer churn.

Recommend the ML problem type,
useful features,
evaluation metrics,
and an ML pipeline.

---

⚙️ Installation

Clone Repository

git clone https://github.com/NikitaR04/crewai-analytics-assistant.git

cd crewai-analytics-assistant

---

Create Virtual Environment

python -m venv .venv

Windows

.venv\Scripts\activate

Linux / macOS

source .venv/bin/activate

---

Install Dependencies

pip install -r requirements.txt

---

🤖 Install Ollama

Download Ollama:

https://ollama.com

Pull a supported model:

ollama pull qwen2.5:7b

or

ollama pull llama3.2:3b

Start Ollama:

ollama serve

---

▶️ Run MCP Server

python -m mcp_server.server

---

▶️ Run the Application

streamlit run app.py

Open the application in your browser:

http://localhost:8501

---

🧪 Running Tests

python tests/test_supervisor.py

python tests/test_analyst.py

python tests/test_scientist.py

python tests/test_csv_profile.py

python tests/test_data_quality.py

python tests/test_sql.py

python tests/test_crew_tools.py

---

🌟 Future Enhancements

• Database connectivity (PostgreSQL / MySQL)
• Interactive data visualization
• PDF report generation
• Authentication & user management
• Conversation memory
• Cloud deployment
• Additional MCP tools
• Support for multiple LLM providers

---

🎯 Skills Demonstrated

• Multi-Agent AI Systems
• CrewAI
• Ollama
• Model Context Protocol (MCP)
• Streamlit Application Development
• Prompt Engineering
• Python Backend Development
• Data Analytics
• Machine Learning Workflow Design
• SQL Validation
• YAML Configuration

---

👩‍💻 Author

Nikita Rani

B.Tech, Computer Science & Engineering
KIIT University, Bhubaneswar

Project developed as part of the IIT-R ML & Agentic AI Summer Internship Training Program.

---

📄 License

This project is intended for educational and portfolio purposes.