Database Driven Teachable Reporting Agent

I need a teachable agent that can ingest live numerical data coming straight from my database records, learn from it, and automatically turn those figures into clear, well-structured reports. The agent’s sole purpose is reporting, not prediction or trend monitoring, so accuracy and readable formatting matter more than fancy forecasting.

Scope of work
• Connect securely to the existing SQL database (read-only access) and pull the required numerical fields.
• Enable incremental learning so the agent stays current as new rows are added.
• Design the reporting logic: aggregate, summarise, and present key metrics with any charts or tables the data justifies.
• Package everything—code, trained artefacts, and a short setup guide—into a single deliverable file ready for hand-off on or before 26 June.

Acceptance criteria
1. Reports reflect the latest database state at run-time.
2. All calculations trace back to the original numeric values without manual intervention.
3. Final file runs locally on a standard Python stack (pandas, SQLAlchemy and any lightweight ML library are fine) with a one-command start.

Feel free to recommend additional libraries if they streamline the workflow, provided the final hand-off stays simple and self-contained.