🧾 ACME Spend Categorization (UNSPSC CodeGenAI)

An advanced, intelligent spend categorization system designed to automatically classify invoice-level procurement data using a hybrid approach that combines both rule-based logic and AI-driven semantic understanding. The system is architected to be scalable, accurate, and human-in-the-loop friendly, enabling organizations to process large volumes of invoice descriptions while maintaining control, auditability, and transparency over categorization decisions.

---

🚀 Overview

This system processes invoice-level data to categorize spend items according to the UNSPSC (United Nations Standard Products and Services Code) taxonomy. It uses a hybrid pipeline that combines:

• AI-Powered Categorization using OpenAI GPT models
• Rule-Based Classification for deterministic tagging
• Manual Review Interface for edge cases and quality assurance

---

✨ Key Features

• Automatic Categorization: Categorize thousands of line items at scale
• Hybrid Classification: Leverages both deterministic rules and probabilistic AI
• Confidence Scoring: Track classification certainty with every result
• Manual Review: Built-in UI for validating uncertain predictions.
• Data Visualization: Interactive spend analytics and can be downloaded as well for detailed analysis.
• Export Capability: Download categorized and reviewed data in CSV format
• Taxonomy-Aware Embeddings: Uses semantic similarity against the UNSPSC hierarchy
• Daily Taxonomy Sync: Refreshes UNSPSC data every 3 months automatically
• Real-time Search: Instant search across categorized items
• Learning System: Use feedback to improve future classifications
• Feedback History: Maintain audit trail of all corrections
• Continuous Improvement: Regular prompt updates based on feedback

---

📊 Enhanced Performance Metrics

• Detailed Analytics Dashboard:
  • Confidence distribution analysis
  • Category-wise performance metrics
  • Source distribution (Rule-based vs AI-based)
• Downloadable Reports:
  • Export metrics as CSV

---

🎯 Prompt Optimization

• Dynamic Prompt Engineering: Automatically optimize prompts based on performance
• Context-Aware Prompts: Adapt prompts based on item category
• Continuous Improvement: Regular prompt updates based on feedback

---

📈 Performance Visualization

• Interactive Charts:
  • Category distribution bar graphs
  • Supplier amount distribution
  • Confidence score pie charts
  • Source distribution analysis
• Downloadable Visualizations:
  • Export charts as PNG
  • Save metrics as CSV
  • Generate comprehensive reports

---

📁 Data Format

Prepare your invoice CSV like this:

Invoice ID	SKU	Description	Supplier	Amount
001	10001	Black toner cartridge	OfficeSupplyCo	89.99
002	20003	Fiber optic cables, 50ft	NetGear Inc.	129.50

Required columns: description, supplier, sku, invoice_id

---

🌐 Web UI

• 👉🏻Check Deployed Site: Live Demo

• Main Dashboard
  
  

• After Running the Model
  
  

• Categorised Table
  
  

• Category Distribution Bar Graph
  
  

• Supplier By Amount Distribution Bar Graph
  
  

• Supplier Distribution Pie Chart
  
  

• Confidence Distribution Pie Chart
  
  

• Manual Review UI
  
  

• Manual Review Dropdown and Seachbar
  
  

• Model Performance Metrics
  
  

• Category Wise Performance
  
  

---

💻 How It Works

• Sanitization: Cleans and normalizes invoice text
• Rule Matching: Applies analyst-defined keyword rules
• Semantic Retrieval: Finds top UNSPSC candidates via embedding similarity
• AI Selection: GPT-4 picks the most likely UNSPSC code
• Confidence Routing: Items below threshold are queued for manual review

---

📊 Performance & Monitoring

• Multi-core parallel processing via multiprocessing
• Confidence-based classification routing
• Logs available in logs/pipeline.log

---

📅 Roadmap

• More robust rule engine (regex, entity recognition)
• Spend analytics dashboard
• RESTful API endpoints for integration
• Scheduled batch job manager
• User-friendly UI

---

🤝 Collaborators

• Komal Meena
• Subhav Jain
• Sidhant Budhiraja
• Prayash Pandey

---

⚙️ Prerequisites

• Python 3.9+
• OpenAI API Key

---

📦 Installation

1. Clone the Repository

   git clone https://github.com/yourusername/acme-spend-categorization.git
   cd acme-spend-categorization
   python3 -m venv venv
   source venv/bin/activate  # macOS/Linux
   # OR for Windows
   venv\Scripts\activate
   pip install -r requirements.txt
   python src/08_pipeline.py

   OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

---

🖥️ Usage

1. Run Full Pipeline - Processes invoices and writes outputs

   python src/08_pipeline.py

   ---

2. Outputs

   • data/categorized.csv: High-confidence auto-tagged items
   • data/manual_review.csv: Items requiring human validation
   • logs/pipeline.log: Detailed logging of categorization events

---

📚 Acknowledgments

• OpenAI for powerful AI APIs
• UNSPSC.org for the classification taxonomy
• Sentence-Transformers for semantic search