Customer data dashboard
A working Python dashboard that cleans and joins six business exports to answer nine questions. Built as a workshop demo; the planned AI narration is still in development.
The problem
Hand a chat model your sales file and ask what is going on. It reads part of the file and answers as if it read all of it. Ask twice and you can get two different answers.
What changed
The working dashboard is prepared for a workshop planned for November 3 and 4, 2026. The next step is to give a model the computed results, have it write a plain-language explanation, and check the explanation against those results.
- My work
- Design, build, and teach
- When
- In development · workshop planned November 2026
- Built with
- Python, Claude, Google Colab
What I built
A Python engine that cleans and joins six exports, answers nine business questions, and builds an HTML dashboard. The demonstration uses a fictional coffee business with generated data. The calculations and dashboard are built; AI narration, mapping unfamiliar exports, and weekly email are not integrated yet.
Working with the model
- Model
- Whatever subscription the room already has, which is the constraint the whole design works inside
- What I told it
- Explain these numbers to the owner. The script computed them; do not recompute anything.
- What I threw out
- Letting the model do the arithmetic. Chat models can now write and run code on an uploaded file, but they write new code every time, with no saved rule for what counts and no record of what they cleaned.
- What broke
- The AI narration step is not integrated yet, so this project does not claim a tested end-to-end AI explanation workflow. The current evidence is the Python engine and the generated dashboard.