Xelly – Excel Automation & Analysis

Xelly is a full-stack web application designed to simplify the cleaning, clustering, analysis, and forecasting of trade and product data stored in Excel or CSV files. It leverages machine learning techniques and time-series forecasting to deliver insights without requiring coding expertise.

Live App: clean-excel.vercel.app

---

Table of Contents

1. Problem Statement
2. Proposed Solution
3. Key Features
   • 1. File Upload and Format Support
   • 2. Cosine Similarity-Based Clustering
   • 3. Data Analysis & Filtering
   • 4. Forecasting
   • 5. Company-wise Analysis Dashboard
4. System Architecture
5. Deployment and Security
6. Installation (Local Development)
7. Tech Stack
8. Author

---

Problem Statement

Consulting firms and analysts frequently deal with messy Excel files containing raw trade data with inconsistent entries, typographical errors, and non-standardized naming conventions. Manual cleaning and clustering is tedious and error-prone, while advanced analysis and forecasting are limited in Excel.

---

Proposed Solution

Xelly automates:

• Data cleaning and standardization
• Clustering similar entries using TF-IDF + Cosine Similarity and Sentence Transformers
• Trade data filtering and interactive dashboards
• Forecasting product prices or quantities using Prophet, Linear Regression, and Polynomial Regression

---

Key Features

1. File Upload and Format Support

• Upload Excel (.xlsx) or CSV (.csv) files via drag-and-drop or file picker.
• Automated data cleaning: trimming spaces, normalizing formats, currency (USD), and unit conversions (e.g., UQC → KG).
• Preview cleaned data and download standardized files.

2. Cosine Similarity-Based Clustering

• Groups similar text entries like “Petroleum coke” and “Petroleum coke bulk”.
• Adjustable similarity threshold.
• Option to accept or reject suggested replacements.

3. Data Analysis & Filtering

• Trade Type Detection (Import/Export).
• Multi-level filtering by city, supplier, HS code, value, year, etc.
• Insights include:
  • Top importer-supplier pairs
  • Market share metrics
  • Unit price trends
  • Importer-supplier heatmaps

4. Forecasting

• Time-series forecasts for columns (e.g., quantity or unit price).
• Models:
  • Linear Regression
  • Polynomial Regression
  • Prophet (Meta)
• Outputs:
  • 2-year forecast (line + bar charts)
  • Model performance metrics (MAE, RMSE, R²)
  • Seasonal patterns and historical trends

5. Company-wise Analysis Dashboard

• Key statistics: transaction count, unique products, time range.
• Quantity and price metrics.
• Product-specific transaction records.

---

System Architecture

• Frontend: React (Vercel hosting).
• Backend: Flask API (Render hosting).
• Core Modules:
  • upload_bp – File upload, cleaning.
  • cluster_bp / cosine_bp – TF-IDF & cosine similarity clustering.
  • filter_bp – Trade data filtering and analysis.
  • company_bp – Company-level metrics.
  • forecast_bp – Prophet-based time series forecasting.

---

Deployment and Security

• Frontend: Hosted on Vercel (HTTPS, TLS, CI/CD).
• Backend: Hosted on Render (HTTPS endpoints, environment variables for sensitive keys).
• CORS Policies & Server Validation for data protection.

---

Installation (Local Development)

# Clone repository
git clone https://github.com/saun09/clean-excel.git

# Backend Setup
cd backend
pip install -r requirements.txt
flask run

# Frontend Setup
cd frontend
npm install
npm run dev