Skip to content

README.md

Latest commit

 

History

History
304 lines (235 loc) · 5.25 KB

README.md

File metadata and controls

304 lines (235 loc) · 5.25 KB

🚀 AI Invoice Intelligence Platform

Python FastAPI React TailwindCSS Ollama LLM AI Powered OCR Status License Contributions

A full-stack AI-powered invoice automation system built using:

  • 🧠 Ollama (Local LLM - Qwen2 / Llama3)
  • ⚡ FastAPI (Backend API)
  • 🎨 React + Tailwind (Frontend SaaS UI)
  • 📊 Recharts (Analytics Dashboard)

💼 Business Value

  • Reduces manual invoice entry time
  • Improves extraction accuracy with AI
  • Eliminates dependency on external APIs
  • Keeps financial data fully local & secure
  • Provides analytics for vendor spending insights

🔥 Features

  • Upload single or multiple invoice PDFs
  • Extract:
    • Vendor Name
    • Invoice Number
    • Invoice Date
    • Total Amount
    • Tax
  • Structured JSON output
  • Automatic confidence scoring
  • Accuracy metric
  • Processing time tracking
  • Logs storage (JSON)
  • CSV export
  • Analytics Dashboard:
    • Revenue trend
    • Vendor revenue
    • Confidence visualization
  • Fully local LLM using Ollama (No OpenAI API)

🔷 Detailed Processing Flow

  1. Step-by-Step Execution
  2. User uploads PDF
  3. React sends POST request
  4. FastAPI stores file
  5. OCR extracts raw text
  6. Text is chunked
  7. Each chunk sent to Ollama
  8. JSON results returned
  9. Merge structured fields
  10. Compute confidence + accuracy
  11. Save to logs
  12. Return response to frontend
  13. UI updates dashboard
User
 │
 ▼
Upload PDF
 │
 ▼
FastAPI Endpoint (/process-invoice)
 │
 ▼
OCR Extraction
 │
 ▼
Text Chunking
 │
 ▼
Ollama LLM Processing
 │
 ▼
JSON Extraction
 │
 ▼
Confidence + Accuracy Calculation
 │
 ▼
Log Storage
 │
 ▼
Frontend Dashboard Update

🏗 Architecture

React (Cloud)
        │
        ▼
API Gateway
        │
        ▼
FastAPI (Dockerized)
        │
        ├── Worker Queue (Celery)
        │
        ├── Redis (Task Queue)
        │
        ├── PostgreSQL (Structured Logs)
        │
        └── Ollama Server (Dedicated GPU Machine)

📸 Product Screenshots

📊 Dashboard Overview

Dashboard Overview

📈 Analytics & Revenue Insights

Dashboard Analytics

📁 Invoice Logs & CSV Export

Invoice Logs

🖥 Run Locally

1️⃣ Start Ollama

ollama serve

2️⃣ Start Backend

cd backend
uvicorn main:app --reload --port 8000

3️⃣ Start Frontend

cd frontend
npm install
npm run dev

📈 Tech Stack

| Layer   | Technology       |
| ------- | ---------------- |
| UI      | React + Tailwind |
| Charts  | Recharts         |
| Backend | FastAPI          |
| OCR     | PyMuPDF          |
| LLM     | Ollama           |
| Model   | Qwen2 / Llama3   |
| Storage | JSON / CSV       |

🟢 Frontend (React + Tailwind)

Responsibilities:

  • Invoice Upload (Single/Batch)
  • Real-time Processing UI
  • Dashboard Analytics
  • Logs Table
  • CSV Export Trigger
  • Confidence Visualization

Runs on:

http://localhost:5173

Communicates via REST API.

🔵 Backend (FastAPI)

Responsibilities:

  • Accept file uploads
  • Store temporary files
  • Extract text using OCR
  • Chunk large documents
  • Send structured prompts to LLM
  • Merge extracted data
  • Compute:
    • Confidence score
    • Accuracy metric
    • Processing time
  • Save logs
  • Serve CSV export

Runs on:

http://localhost:8000

🟣 OCR Layer

Technology:

PyMuPDF (fitz)

Purpose:

  • Extract text from PDF pages
  • Handle multi-page documents
  • Provide raw text to LLM

🔴 LLM Layer (Ollama)

Technology:

  • Qwen2:1.5b (fast extraction)
  • Llama3:8b (higher reasoning)

Responsibilities:

  • Structured field extraction
  • JSON-only response
  • Invoice understanding
  • Vendor detection

Runs locally:

ollama serve

🟡 Processing Engine

Inside extractor:

  • Smart chunking (for large invoices)
  • Per-chunk extraction
  • Result merging
  • Weighted confidence scoring

🟤 Storage Layer

Currently:

  • JSON logs file
  • CSV export

📌 Future Improvements

  • WebSocket live progress tracking
  • Line-item extraction
  • Vendor auto-classification
  • Multi-user authentication
  • Cloud deployment option
  • PostgreSQL
  • MongoDB
  • Vector DB (for semantic invoice search)

🎯 Goal

Automate invoice data entry for accounting teams using local AI.

🎯 Target Users

  • Accounting Firms
  • SMEs managing vendor invoices
  • Finance Teams
  • Freelance Bookkeepers
  • Businesses handling bulk PDF invoices