Analyst personas built with prompts

Explore how far prompts alone can shape an assistant for football analysis.

The same model should behave differently when it assesses a player's performance, a transfer proposal or a tactical reading, and each task needs its own rules about what counts as evidence.

A terminal chatbot in TypeScript sends messages to the OpenAI API and streams the answer. Three personas carry their own instructions: sports analysis, transfer-proposal analysis and tactical analysis. The conversation is stored in SQLite, with utilities to limit or summarize the history, and simple checks screen input and output.

No formal evaluation was done in this experiment: the checks are textual heuristics and the repository has no test suite. The same personas are part of Football Stat Analytics, where validation and measurement were built.

The checks do not guarantee that an answer is correct or safe. The repository has no install or run instructions yet, so the experiment cannot be reproduced from it, and it is separate from the future FintaLab product.

Without tests or measurements there is no way to tell whether a persona follows its instructions, which is the gap the validation and evaluation experiments address.