Skip to main content
Version: Next

AI Operator Guide

Updated 2025.03.20

What is an AI Operator?​

An AI Operator is a professional who values field experience and practical know-how in industrial settings and is highly interested in improving product quality, productivity, and operational efficiency. While they may not be AI experts, they possess a basic understanding of data analysis and AI technologies. The primary goal of an AI Operator is to adopt AI technologies to enhance work efficiency and productivity, while ensuring sustained performance at minimal cost.

Challenges Faced by AI Operators​

AI Operators often face challenges such as lack of expertise in building complex AI models, difficulty in collecting, cleaning, processing, and analyzing on-site data, limited budgets for adopting AI technologies, and difficulty in securing experts for ongoing AI operations. Mellerikat provides various features to help overcome these challenges and support AI Operators in achieving their goals.

Automated Solutions with Mellerikat​

To help AI Operators reach their goals, Mellerikat provides tools such as Edge Conductor and model management systems, allowing for easy deployment and operation of AI models. AI Operators can install AI models on various edge devices, collect and validate field data, detect and handle model errors, and more. Using the dataset features in Edge Conductor, they can generate training datasets or improve model accuracy by re-labeling inference results. AI models can also be registered and tested for performance in real-world environments through AI Conductor.

Mellerikat’s Edge Conductor supports deployment of AI models into production environments and enables the preparation of training datasets for model training. Operators can upload datasets or select them from cloud storage, monitor inference results in real-time, and validate outputs to fine-tune models. When inference errors occur, retraining can be performed to improve model performance.

Through this, AI Operators can use Mellerikat to manage and operate AI models at a higher level, contributing real value to business operations with AI. The platform’s features play a critical role in helping AI Operators meet their goals while effectively resolving challenges.

User Manual for AI Operators​

The following subpages provide detailed guidance for AI Operators to use Mellerikat effectively, including specific instructions for required functions and workflows. This enables AI Operators to fully leverage the platform to achieve their objectives quickly and accurately.


Mellerikat Workflow for AI Operators​

  1. Set Up Environment: The AI Operator installs Edge App on the target edge device, PC, or cloud environment where the AI model will run. Edge Conductor is installed on a server or cloud instance and configured to connect with Edge App. Setup requires configuration of connection details with Edge Conductor.

  2. Register Edge: Within Edge Conductor, the AI Operator checks available edge devices and registers the ones they wish to use.

  3. Create Stream: The AI Operator creates a Stream and selects an AI Solution instance. User-defined parameters can be set for the AI Solution to reflect operational preferences.

  4. Generate Training Dataset: After creating the Stream, a dataset for training the AI model is prepared. Depending on its location, the dataset source can be an edge device, local storage, or S3.

  5. Train Model: The training process is initiated through the Stream using the selected dataset. A model training request is sent to AI Conductor, and the trained model is returned upon completion. All training history is managed within the Stream.

  6. Deploy Model: Once training is complete, the AI model is deployed to an edge device. The operator selects which registered edge to deploy to. Previously trained models can also be redeployed.

  7. Manage Inference Results: Inference results from edge devices are sent to Edge Conductor, which collects and manages them.

  8. Rebuild Dataset & Retrain: Based on collected inference results, a new training dataset can be built. If errors are identified in the inference outputs, re-labeling can be performed to improve data quality. The updated dataset can then be used to retrain and optimize the AI model.


Topics​