Methodology — Automated Intelligence Reporting (LLM)

The methodology treats LLM output as a first draft that must remain grounded in verified indicator data.

1. Data validation

Start with structured indicators, targets, actuals, risk notes, and recommendations.

2. Narrative planning

Map indicator status to report sections such as progress, risks, mitigation, and next steps.

3. Controlled generation

Generate concise summaries from only the available demo evidence.

4. Human review

Require M&E or programme review before donor submission.

Evaluation approach

Responsible AI note: The sandbox demonstrates assistive decision support. In production, human review, audit logging, access control, model monitoring, and documented exception handling would be required.