🧾 AI Receipt Parser

A full-stack AI application that converts unstructured receipt images into structured JSON data using Groq vision API.

🚀 Live Demo

• Frontend: [here]

• Backend API Docs: [here]

AI-Powered Receipt Extraction

This project is an AI-powered extraction system. It solves the problem of manual data entry by allowing users to upload a photo of a receipt and instantly getting back structured data (Store Name, Date, Total, Tax, Line Items).

It uses Groq's llama-4-scout Vision model to scan the image and Pydantic to strictly validate the output, ensuring the data is clean and ready.

✨ Key Features

• 👀 Visual Intelligence: Uses Large Multi-Modal Models (LMMs) to read text from images, handling complex layouts better than traditional OCR.

• 🛡️ Robust Validation: Implements strict Pydantic models. If the AI hallucinates a date format, the backend catches and corrects it (or flags it) before it reaches the UI.

• ⏱️ Custom Rate Limiting: Features a custom-built, sliding-window rate limiter (Async/No-Dependency) to protect the API from abuse and manage 3rd-party token limits.

• 🐳 Fully Containerized: Split into Microservices (Frontend & Backend) orchestrated via Docker Compose.

• 📊 Interactive Dashboard: A Streamlit UI that visualizes extraction confidence, tax breakdowns, and itemized lists in real-time.

🛠️ Tech Stack

Backend

• Framework: FastAPI
• Language: Python 3.11
• AI Provider: Groq SDK (llama-4-scout)
• Validation: Pydantic

Frontend

• Framework: Streamlit

🚀 Getting Started

Prerequisites

• Docker installed
• A Groq API Key (Get one at Groq)

1. Clone the Repository**

git clone https://github.com/Aman00240/RECEIPT-PARSER
cd RECEIPT-PARSER

2. Configure Environment

Create a .env file in the root directory:

GROQ_KEY=groq_api_key
MODEL=meta-llama/llama-4-scout-17b-16e-instruct

3. Run with Docker Compose

This builds both services and sets up the internal network.

docker-compose up --build

• Frontend: Open http://localhost:8501

• Backend Documentation: Open http://localhost:8000/docs

🏗️ Architecture Flow

graph LR
    User[User Uploads Image] --> Frontend(Streamlit UI)
    Frontend -- HTTP POST --> Backend(FastAPI)
    Backend -- Async Check --> Limiter{Rate Limit?}
    Limiter -- Yes --> Error[429 Too Many Requests]
    Limiter -- No --> AI_Service[Groq AI Service]
    AI_Service -- Vision Analysis --> LLM(llama-4-scout)
    LLM -- Raw JSON --> Validator(Pydantic Schema)
    Validator -- Structured Data --> Backend
    Backend -- JSON Response --> Frontend
    Frontend -- Display Metrics --> User

Loading

📂 Project Structure

.
├── app/
│   ├── routers/
│   │   └── receipts.py      # API Endpoints
│   ├── services/
│   │   └── vision.py        # Groq Integration & Prompt Engineering
│   ├── config.py            # Environment Settings
│   ├── main.py              # FastAPI Entrypoint
│   ├── schemas.py           # Pydantic Data Models
│   └── utils.py             # Rate Limiter Logic
├── frontend/
│   └── ui.py                # Streamlit Dashboard
├── Dockerfile               # Backend Image
├── Dockerfile.frontend      # Frontend Image
├── docker-compose.yml       # Orchestration
└── requirements.txt         # Dependencies for backend image
└── requirements_frontend.txt # Dependencies for frontend image

🔌 API Usage

The API comes with interactive documentation via Swagger UI. Once running, navigate to: http://localhost:8000/docs

Quick CLI Test

You can also test the endpoint directly from your terminal:

curl -X POST "http://localhost:8000/receipts/scan" \
     -H "accept: application/json" \
     -H "Content-Type: multipart/form-data" \
     -F "file=@/path/to/receipt.jpg"

Response Example:

{
  "store_name": "Store",
  "total_amount": 14.50,
  "tax_amount": 1.20,
  "scan_confidence":1,
  "items":[
    {"item_name": "Milk", "price": 4.50}
  ]
}