VizData AI: Intelligent Spreadsheet Operations & Analytics

VizData AI is a powerful analytical platform that integrates autonomous AI agents directly into an interactive spreadsheet interface. By leveraging Local LLMs (Ollama) and the CrewAI multi-agent framework, users can chat directly with their data—telling the AI to write Pandas scripts to generate test datasets, modify spreadsheets, or automatically plot deep analytical visualizations, all completely offline.

---

Key Features

• Dual-Agent Architecture:
  • Data Manager: Executable Data Science agent that natively writes pandas Python code to build, edit, sum, or reshape data interactively in real-time.
  • Analytics Agent: Visual statistician that ingests spreadsheet state to automatically generate and display complex matplotlib and seaborn graphs directly in your chat.
• Spreadsheet Aesthetic: A fully interactive grid (built entirely in Vanilla JS/CSS) featuring modern glassmorphic styling, collapsible chat sidebars, and real-time editable cells.
• Privacy-First Local LLMs: Built to run entirely locally with Ollama (using models like Qwen 2.5), overriding standard OpenAI parsers to enforce strict API tool-calls natively.
• Multi-Format Uploads: Effortlessly inject .csv, .xlsx, and .xls files directly into the active pandas memory state.

---

Tech Stack

Layer	Technologies
Frontend	Vanilla HTML5, Advanced CSS3, Vanilla JS, Lucide Icons
Backend	FastAPI, Uvicorn, Pandas, Matplotlib, Seaborn
AI "Brain"	CrewAI, LiteLLM, Ollama (Local), Google Gemini (Cloud)
Environment	Python-dotenv

---

Architecture

graph TD
    User((User)) -->|Upload Data| UI[Frontend UI]
    User -->|Chat Query / Create| UI
    UI -->|API Payload| Backend[FastAPI Server]
    
    subgraph "Intelligent Multi-Agent System"
        Backend -->|Natural Language| CrewAI
        CrewAI -->|Routing| Router{Query Type?}
        
        Router -->|CRUD & Generation| DataAgent[Data Manager]
        DataAgent -->|Writes & Runs Pandas Code| PyTool[run_python_script]
        PyTool -->|Mutated DataFrame| RAM[(Active Session RAM)]
        
        Router -->|Visualization| VizAgent[Analytics Agent]
        VizAgent -->|Creates PNGs| ChartTool[Plotting Tools]
    end
    
    RAM -->|Updated JSON State| Backend
    ChartTool -->|Base64 Image Array| Backend
    Backend -->|Refresh UI & Chat| UI

Loading

---

Getting Started

1. Prerequisites

• Python 3.9+
• Ollama (If running locally, pull qwen2.5:7b or similar)
• (Optional) Gemini API Key for cloud fallback.

2. Installation

Clone the repository:

git clone https://github.com/Shaktisinhchavda/vizdata-ai.git
cd vizdata-ai

Set up the Environment:

# Create a virtual environment
python -m venv venv

# On Windows:
venv\Scripts\activate
# On MacOS/Linux:
source venv/bin/activate

Install Backend Dependencies:

pip install -r requirements.txt

Environment Variables: Create a .env file in the root directory if you want to fall back to Google Gemini models instead of locally emulated Ollama paths:

GEMINI_API_KEY=your_gemini_api_key_here

(Note: If no key is provided, the FastAPI backend falls back recursively to local http://localhost:11434/v1 OpenAI-emulated endpoints for total offline security)

---

How to Use

1. Start the Server:

   cd backend
   python main.py

2. Access the App: Open http://localhost:8000 in your browser.
3. Analyze Data:
   • Upload a dataset or click Create with AI to let the Data Manager generate a dataset from scratch using a prompt.
   • Example prompt for Data Manager: "Create an empty dataset with columns Name, Age, Title, and Salary. Then add 5 rows of sample employee data."
   • Example prompt for Analytics Agent: "Plot a scatter matrix showing the relationship between Revenue and Units Sold grouped by Region."

---

License

Distributed under the MIT License. See LICENSE for more information.