Skip to main content

[Maeumjigi Capstone Dev Log #3] Alarm App Build Test and Fine-Tuning Preprocessing Kickoff

· 2 min read
Jihwan Park (jaybee)
Jihwan Park (jaybee)
AI & Robotics Engineer

Cover Image

This is the record of March 31, when we conducted the Android build test of the MND Convention reservation alarm app, which was the initial direction of the Maeumjigi project, while simultaneously starting the Llama 3.1 fine-tuning data preprocessing, hinting at a full transition into an AI-based project.

Android App Build and Unit Testing​

  • Problem: We needed to verify if the server and Android client codes planned and generated yesterday would successfully build and pass tests in the actual Android Studio environment.
  • Process:

    To Claude Code: "Current progress... Stop all work... What were you working on?"

  • Result: Numerous build artifacts and metadata were generated in the mnd-reservation-alarm/android/app/build/... path. We executed unit tests such as NotificationBuilderTest and checked the Android-specific build pipeline, including debug resource merging and manifest processing. The core logic of detecting reservation changes in the background and popping up alarms was tested.

The Great Turning Point: Starting Llama 3.1 Fine-Tuning Preprocessing​

  • Problem: Moving beyond a simple alarm app, preparation was needed to integrate an advanced AI that understands the user's (soldier's) reservation context and emotional state.
  • Process:

    To Claude Code: "Please proceed with the work according to llama31_finetuning_data_preprocessing_plan_short.md" "No additional permission is needed for any choices regarding this task."

  • Result: We took a crucial first step that would completely change the nature of the project. We instructed the work according to the data preprocessing plan for fine-tuning the Llama 3.1 model. To speed up the process, we granted the AI strong authority by saying, "execute immediately without asking for additional permission for all choices," and started the massive data processing pipeline.

Terminal Screenshot

This day was a meaningful day when the Android build work to improve the completeness of the alarm app and the AI fine-tuning data preprocessing, which will become the core feature of the future, started in parallel. From now on, the arduous preprocessing journey of processing tens of thousands of conversation data points to fit Llama 3.1 will begin in earnest!