Wed, Oct 08, 25, DEVELOPMENT ISSUES - Auto-imported from uconGPT project

Auto-imported from: D:/repos/aiegoo/uconGPT/eng2Fix/kor2fix/DEVELOPMENT-ISSUES.md
Original filename: DEVELOPMENT-ISSUES.md
Import date: Wed, Oct 08, 25

kor2Unity Development Issues

🎯 Current Development Focus

Korean language learning platform with self-hosted LLM integration

📋 Open Issues Summary

High Priority Issues

Issue #1: LLM Integration Strategy

  • Status: 🔄 In Progress
  • Priority: High
  • Focus: Self-hosted model integration with existing infrastructure
  • Key Decisions:
    • Primary environment: minigpt4 conda environment
    • Model choice: Llama 2 7B-HF with MiniGPT-4 multimodal capabilities
    • Port allocation: Migrate Ollama service to port 8203

Issue #2: UI Architecture Decision - TUI vs Web UI

  • Status: 🤔 Decision Required
  • Priority: High
  • Focus: Choose between Terminal UI and Web UI for Korean learning
  • Current Recommendation:
    • Phase 1: TUI MVP leveraging existing infrastructure
    • Phase 2: Web UI for broader user adoption
  • Decision Matrix Score: Web UI (38/50) vs TUI (35/50)

Issue #3: Conda Environment Configuration

  • Status: ⏳ Ready to Start
  • Priority: High
  • Focus: Activate and validate minigpt4 environment for model deployment
  • Next Action: conda activate minigpt4 and environment validation

Infrastructure Status

✅ Completed

  • FastAPI backend foundation (/home/hsyyu/llm_api.py)
  • Docker service orchestration (MongoDB, Ollama, backend)
  • Dedicated port allocation strategy (8200-8299 range)
  • Model inventory and hardware compatibility assessment

🔄 In Progress

  • Conda environment activation for model integration
  • TUI vs Web UI architecture decision
  • Korean language learning feature specification

📋 Planned

  • Model loading and inference pipeline
  • Korean tokenization and prompt engineering
  • Frontend interface implementation (TUI or Web)
  • Integration testing and validation

🎛️ Technical Architecture

Current Stack

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Frontend      │    │   Backend       │    │   AI Models     │
│   (TUI/Web)     │────│   FastAPI       │────│   minigpt4      │
│   Port: 8200    │    │   Port: 8201    │    │   + Llama2-7B   │
└─────────────────┘    └─────────────────┘    │   Port: 8203    │
                                              └─────────────────┘
          │                       │                       │
          │                       │                       │
┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Database      │    │   Services      │    │   Environment   │
│   MongoDB       │    │   Docker        │    │   Conda         │
│   Port: 8202    │    │   Compose       │    │   minigpt4      │
└─────────────────┘    └─────────────────┘    └─────────────────┘

Available Models

  • Llama 2 7B-HF: /home/hsyyu/llama2-7b-hf/ (Machine appropriate)
  • MiniGPT-4: /home/hsyyu/checkpoints/ (Multimodal capabilities)
  • Model Size: 277MB pretrained + 47MB stage2 checkpoints

Development Environments

  • Primary: minigpt4 conda environment (recommended)
  • Fallback: ai-cplus-dev conda environment
  • Future: llama4-env (requires high-end hardware)

🚀 Development Roadmap

Week 1: Environment & Model Setup

  1. Activate minigpt4 conda environment
  2. Validate model loading and inference
  3. Test Korean text processing capabilities
  4. Benchmark performance on current hardware

Week 2: UI Decision & Implementation

  1. Finalize TUI vs Web UI architecture decision
  2. Begin frontend implementation based on choice
  3. Integrate Korean IME and text input handling
  4. Create basic Korean learning interface

Week 3: Backend Integration

  1. Extend FastAPI for Korean learning endpoints
  2. Implement model inference pipeline
  3. Add Korean tokenization and prompt engineering
  4. Create learning session management

Week 4: Testing & Validation

  1. Integration testing across full stack
  2. Korean language learning feature validation
  3. Performance optimization and tuning
  4. Documentation and deployment preparation

📊 Success Metrics

  • Successful conda environment activation
  • Model loading and inference validation
  • Korean text processing accuracy
  • UI responsiveness and usability
  • End-to-end learning experience testing

🔄 Next Immediate Actions

  1. Environment Activation: conda activate minigpt4
  2. Model Validation: Test model loading capabilities
  3. UI Decision: Choose TUI vs Web UI approach
  4. Development Sprint: Begin implementation based on decisions

Last Updated: August 7, 2025 Repository: aiegoo/kor2unity Branch: environment-setup-v1