Skip to main content
Version: docs v25.02

AI Solution Development

ALO Installation​

ALO is a core framework for AI solution development. It enables developers to design and test models, and easily register them on the Try platform for use in operational environments.

Preparing the Development Environment​

Set up a development environment based on Python 3.10. Refer to the Development Environment Setup Guide for detailed instructions.

  • Recommended Environments: Linux or Windows
  • Alternative: If a local environment isn’t suitable, use a cloud platform like Google Colab.

Complete the preparation by installing required packages and configuring environment variables for ALO usage.

ALO Quick Run​

Follow the ALO Quick Run guide to install ALO and run the provided examples. This is the first step to confirm that ALO is functioning correctly.


AI Solution Development​

Developing an AI solution in Mellerikat involves three key steps: Problem Definition, Data Preparation, and Model Code Development. Each step supports efficient development through a structured approach.

Problem Definition​

Problem definition is the critical starting point that sets the direction for your AI solution.

  • Specify the Problem: Clearly define the task to address, e.g., "product defect detection" or "customer satisfaction prediction."
  • Define Input/Output: Specify the input data (e.g., tabular sensor data) and output results (e.g., defect status).
  • Select Performance Metrics: Choose evaluation criteria such as accuracy, precision, recall, or F1 score.

Clarifying goals and scope at this stage makes subsequent development much smoother.

Data Preparation​

Prepare the data needed for model training and operation.

  • Training Data: Collect labeled datasets for model training. Data quality and quantity significantly impact performance.
  • Inference Data: Prepare input data for the model to process in an operational environment, reflecting real-world use cases.
  • Data Preprocessing: Optimize data by handling missing values, normalization, or format conversion.

If data is scarce or misaligned, consider using Mellerikat’s sample datasets.

Model Code Development​

Mellerikat supports tailored model development based on user needs. Below are approaches depending on your situation.

- When Model Development Is Needed​

If you’ve defined the problem and data but haven’t found a suitable model, explore solutions via Mellerihub.

  • Mellerihub: Provides ALO-based AI solutions for domain-specific problems in a Git Repository. Users can reference these to select or customize a model for their needs.

Mellerihub’s AI Contents offers tutorials to create models with simple parameter adjustments:

These tutorials include step-by-step code and explanations, making them beginner-friendly.

Note: The GitHub token provided during the preparation phase will be used to clone Mellerihub's private repository. When executing the 'git clone' command, use your GitHub ID as the username and enter the provided token instead of the password.

- Converting Existing Code to ALO​

To convert an already developed model into ALO format:

The converted solution can be registered in AI Conductor for testing and deployment in an operational environment.


AI Solution Registration​

Once the completed AI solution is registered with AI Conductor, it can be utilized in the operational environment. AI Conductor offers version control for AI solutions, model training using cloud resources, resource usage monitoring, and log recording. The Train Pipeline of the AI solution operates with datasets provided by Edge Conductor to generate the model.

Preparation Steps​

Set up the AWS environment for registration:

  • AWS CLI Installation: Install AWS CLI following the guide for your operating system.
  • ALO Infra Setup: Create an AWS Profile with your issued AWS Key and write the solution_info.yaml file.
    • AWS Profile Setup: Use the aws configure command to input your Access Key and Secret Key.
    • solution_info.yaml Writing: Specify the solution name, version, data paths, etc.

Registration Verification​

Check the AI Conductor dashboard to ensure the solution is registered correctly. You can write detailed information about the AI solution or check the model's training and resource usage.

Solution Register Screenshot