Festiscan Backend

FertiScan helps inspectors analyze and process fertilizer labels by extracting text and generating structured forms.

Overview

This repository contains the backend for FertiScan, a FastAPI-based server designed to work with the frontend. It handles image uploads, document analysis using OCR, and form generation using an LLM.

Setup for Development

Prerequisites

• Python 3.11+
• pip
• virtualenv
• Azure Document Intelligence and OpenAI API keys

Running Locally

1. Clone the repository:

   git clone https://github.com/ai-cfia/fertiscan-backend.git
   cd fertiscan-backend

2. Install dependencies:

   pip install -r requirements.txt

3. Start the server in development mode:

   fastapi dev app/main.py --port 5000

Running with Docker

1. Build the Docker image:

   docker build -t fertiscan-backend \
   --build-arg ARG_AZURE_API_ENDPOINT=your_azure_form_recognizer_endpoint \
   --build-arg ARG_AZURE_API_KEY=your_azure_form_recognizer_key \
   --build-arg ARG_AZURE_OPENAI_DEPLOYMENT=your_azure_openai_deployment \
   --build-arg ARG_AZURE_OPENAI_ENDPOINT=your_azure_openai_endpoint \
   --build-arg ARG_AZURE_OPENAI_KEY=your_azure_openai_key \
   --build-arg ARG_FERTISCAN_STORAGE_URL=your_fertiscan_storage_url \
   --build-arg ARG_FERTISCAN_DB_URL=your_fertiscan_db_url \
   --build-arg ARG_FERTISCAN_SCHEMA=your_fertiscan_schema \
   --build-arg ARG_ALLOWED_ORIGINS=["http://url.to_frontend/"] \
   --build-arg OTEL_EXPORTER_OTLP_ENDPOINT=your_phoenix_endpoint \
   --build-arg ARG_PROMPT_PATH=path/to/file \
   --build-arg ARG_UPLOAD_PATH=path/to/file \
   .

2. Run the Docker container:

   docker run -p 5000:5000 fertiscan-backend

Docker Compose

1. Create a .env file from .env.template. Include the following environment variables:

   FERTISCAN_DB_URL=postgresql://postgres:postgres@postgres:5432/fertiscan
   BB_URL=bytebase_url
   BB_SERVICE_ACCOUNT=your-bytebase-sa@service.bytebase.com
   BB_SERVICE_KEY=your-bytebase-sa-key
   BB_INSTANCE_ID=your-bytebase-instance-id
   BB_DATABASE_ID=your-bytebase-database-id

   You can find their values in our vault under fertiscan-dev.

2. Start the Docker container:

   docker-compose up --build

Side note: If you are on an ARM-based machine, you will need to build the image with the docker-compose build --build-arg TARGETARCH=arm64 command.

The application will be available at http://localhost:80. The database should be dynamically built based on the latest schema from Bytebase.

To use pgAdmin, navigate to http://localhost:5050 and log in with admin@example.com and admin. You can then register a new server with the following details:

• Host: postgres
• Port: 5432
• Username: postgres
• Password: postgres
• Database: fertiscan

Environment Variables

Create a .env file from .env.template.

AZURE_API_ENDPOINT=your_azure_form_recognizer_endpoint
AZURE_API_KEY=your_azure_form_recognizer_key
AZURE_OPENAI_API_ENDPOINT=your_azure_openai_endpoint
AZURE_OPENAI_API_KEY=your_azure_openai_key
AZURE_OPENAI_DEPLOYMENT=your_azure_openai_deployment

FERTISCAN_DB_URL=your_fertiscan_db_url
FERTISCAN_SCHEMA=your_fertiscan_schema

UPLOAD_PATH=path/to/file
ALLOWED_ORIGINS=["http://url.to_frontend/"]

API Endpoints

The Swagger UI for the API of FertiScan is available at /docs.

More details in the developer documentation.