ProcessumAir

Autonomous Data Cleaning & Feature Engineering Agent

ProcessumAir is a next-generation Data Engineering Agent designed to democratize the data preparation process. Powered by Google Gemini 2.5 Flash, it acts as an intelligent autonomous loop that ingests raw datasets (CSV/Excel), identifies quality issues, creates a rigorous cleaning strategy, and executes it to produce Machine Learning-ready data.

ProcessumAir is an autonomous agent that inspects, cleans, and engineers features for your datasets. From raw CSV to ML-ready code in seconds.

🚀 Key Features

• Autonomous Reasoning: Analyzes dataset schema and user goals (e.g., "Predict Churn") to formulate a tailored cleaning plan.
• Universal Import: Drag-and-drop support for CSV and Excel (.xlsx) files.
• Smart Profiling: Automatic detection of data types, missing values, and outliers.
• Trace Visibility: A visual "Decision Trace Graph" showing the agent's internal thought process (Input → Reasoning → Code).
• Client-Side Processing: Data parsing and heuristic cleaning happen locally in the browser for speed and privacy.
• Deliverables:
  • Cleaned Dataset: Download the processed file as .xlsx.
  • Python Script: Get a reproducible Pandas/Scikit-Learn script.
  • Executive Audit Report: A professional PDF certificate of data health.

🛠️ Tech Stack

• Frontend: React, TypeScript, Tailwind CSS
• AI Engine: Google Gemini API (@google/genai)
• Data Engine: SheetJS (xlsx)
• Visualization: Recharts
• Reporting: jsPDF

📦 Installation & Setup

1. Clone the repository

   git clone https://github.com/yourusername/processum-air.git
   cd processum-air

2. Install dependencies

   npm install

3. Configure API Key ProcessumAir requires a Google Gemini API Key.

   Create a .env file in the root directory:

   # Note: In a Vite setup, you might need to configure 'define' in vite.config.ts 
   # to support 'process.env.API_KEY' or switch to 'import.meta.env'.
   API_KEY=your_google_gemini_api_key_here

4. Run the development server

   npm run dev

📂 Project Structure

processum-air/
├── src/
│   ├── components/       # UI Components (Dashboard, Charts, Modals)
│   ├── services/         # Gemini AI Service & Prompt Engineering
│   ├── utils/            # CSV/Excel Parsers & Cleaning Logic
│   ├── types.ts          # TypeScript Interfaces
│   └── App.tsx           # Main Application State & Routing
├── public/
└── package.json

🧠 System Architecture

1. Profiling Phase: The app uses SheetJS to parse the file locally. It generates a statistical metadata summary (Row count, Null %, Types) without sending raw rows to the cloud.

2. Planning Phase: This metadata + User Goal is sent to Gemini 2.5 Flash. The LLM returns a JSON Cleaning Plan (List of steps with reasoning and Python code).

3. Execution Phase: The dashboard simulates the execution. The cleanAndExportData utility replicates the logic (Imputation, Dropping columns) on the full dataset in the browser memory.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Verified by ProcessumAir