Selected case studies
These short case studies explain what each project tried to do, how it was built, and what it cannot prove. They are evidence of working methods, not claims that a personal project is production advice.
LLM intelligence and cost
- Purpose: make model capability and price trade-offs easier to inspect.
- Data and method: public model evaluations and pricing are assembled into a visual comparison, with the refresh process kept in the project repository.
- Limitation: benchmark scores are not a complete measure of usefulness for a particular workflow.
- Learned: a cost-performance chart is most useful when its version, source, and update date remain visible.
- Open the LLM model analysis project →
Sumo Elo Explorer
- Purpose: compare strength across sumo eras with a consistent rating method.
- Data and method: historical bout records are transformed into Elo ratings and exposed through an interactive explorer.
- Limitation: ratings depend on the data window and model assumptions; they are not a definitive ranking of wrestlers.
- Learned: making assumptions visible matters as much as making the chart attractive.
- Open the Sumo Elo project →
Hong Kong Rent Index
- Purpose: turn public housing and rent data into a readable local indicator.
- Data and method: public data is cleaned, indexed, and presented with the source and calculation context.
- Limitation: an index describes a selected data series; it does not represent every household or property.
- Learned: a small, inspectable data product can be more useful than a broad but opaque dashboard.
- Open the Hong Kong Rent Index →