Timeline
Chronological progression of work from October 2024, documenting key decisions, pivots, and migration milestones in the optimization of Ghana educational apps.
Identified that apps are too pricey. Intervention Courses - Teacher Planning app is the most used, currently using Claude Sonnet 4.5. Decision made to keep Sonnet 4.5 for this critical app given importance, or potentially move to cheaper model.
Key Decisions:
- Intervention Courses app stays on Sonnet 4.5 given importance
- Skip Ghana optimization for now - data was from testing, not real-world usage
Sonnet models hitting rate limits and high costs. Goal: get as many apps as possible to Kimi K2 given best cheap tool calling ability. Started reviewing Sonnet conversations and testing same conversations with Kimi.
Costs became unbearable. Doubling down on moving all apps to Kimi K2 and improving tool calling. Also addressing issue with Claude Sonnet constantly providing assessments with question B being the main answer.
Moving all apps from current reference structure to Teaghan's general .md file structure. New plan focuses on: cost cutting experiments with lower cost models, testing adding more inputs to reduce conversation length, optimizing workflow to generate lesson plan only once.
Key Decisions:
- Undo all reference modifications done previously
- Focus on cost cutting and workflow optimization
Migrated apps from traditional RAG method to 'Search References Tool'. Updated multiple subject-specific apps including Physics, Performing Arts, PEH, Food and Nutrition, Clothing and Textiles, Government, Geography, Biology, and many Ghanaian Language apps.
Testing Intervention, English, Math, and Economics apps across Qwen, Gemini 2.5 Flash: Preview, and Kimi K2. Added disclaimer that MCQs must be double checked. Subject choice requirement removed. Gemini disqualified due to poor tool calling.