🧾 AI Invoice Analyzer

---

🚀 Project Overview

AI Invoice Analyzer is an end-to-end document intelligence prototype built in Google Colab that extracts structured invoice data from images and PDFs using OCR + a local Large Language Model.

This project demonstrates how traditional OCR pipelines can be augmented with LLM reasoning to convert noisy, unstructured documents into validated, machine-readable JSON — a common real-world problem in finance and operations.

⚠️ Current status: Notebook-based prototype (Colab) 🚧 Designed with a clear path toward productionization

---

🧩 What Problem Does This Solve?

Manual invoice processing is:

• Time-consuming

• Error-prone

• Expensive at scale

This system automates:

• Invoice data extraction

• Structural normalization

• Logical validation (totals, dates)

All without relying on cloud LLM APIs, making it suitable for privacy-sensitive environments.

---

🧠 Key Capabilities

📄 Accepts invoice images and PDFs

🔍 OCR extraction using Tesseract

🧠 Local LLM (Ollama) for structured JSON extraction

📋 Extracted fields:

• Vendor name

• Invoice number

• Invoice date & due date

• Subtotal, tax, total

• Currency

✅ Automatic validation:

• Total ≈ Subtotal + Tax

• Due date > Invoice date

• Invoice number format checks

🌐 Interactive Gradio web interface

🔐 Fully local LLM inference (no OpenAI / cloud APIs)

---

🏗️ System Architecture (Conceptual)

Invoice (PDF / Image)
        │
        ▼
 Image Preprocessing
 (Resize, optimize)
        │
        ▼
   OCR Engine
 (Tesseract)
        │
        ▼
 Extracted Raw Text
        │
        ▼
 Local LLM (Ollama)
 (Prompt-engineered
 JSON extraction)
        │
        ▼
 Structured Invoice Data
        │
        ▼
 Business Rule Validation
        │
        ▼
 JSON Output + UI Display

---

🛠️ Tech Stack

Category	Tools
Language	Python
Notebook	Google Colab
OCR	Tesseract
LLM	Ollama (LLaMA 3.2 / Phi-3 / Gemma)
PDF Handling	pdf2image
Image Processing	Pillow
UI	Gradio
Validation	Custom rule-based logic

---

▶️ How to Run (Google Colab)

1️⃣ Open the notebook:

notebook/AI_Invoice_Analyzer.ipynb

2️⃣ Run cells top to bottom:

• Installs system dependencies

• Starts Ollama

• Pulls LLM model

• Launches Gradio UI

3️⃣ Upload an invoice:

• File types: .jpg, .png, or .pdf

4️⃣ View:

• Structured summary

• Full validated JSON output

⚠️ First run may take 30–60 seconds due to model loading.

---

📊 Example Output

{
  "vendor_name": "ABC Supplies Ltd",
  "invoice_number": "INV-2024-001",
  "invoice_date": "2024-10-01",
  "due_date": "2024-10-30",
  "subtotal": 950.0,
  "tax": 50.0,
  "total": 1000.0,
  "currency": "USD"
}

---

🎥 Demo Video:

👉 https://drive.google.com/file/d/1sXG0nK2BgfmmM2CFF-Ne70hKstxMJmzv/view?usp=sharing

---

🧠 Engineering & Learning Highlights

• Designed a modular AI pipeline inside a notebook:

  • Preprocessing

  • OCR

  • LLM extraction

  • Validation

• Prompt-engineered strict JSON-only LLM responses

• Implemented confidence scoring and validation logic

• Managed local LLM lifecycle inside Colab

• Focused on production constraints:

  • Latency

  • Validation

  • Error handling

  • Privacy

---

🔮 Roadmap to Production

This notebook is intentionally structured to be split into modules:

• Refactor into Python packages

• Expose FastAPI REST endpoints

• Add batch invoice processing

• Dockerize OCR + LLM services

• Persist results to a database

• Deploy on VM / on-prem environment

---