15 · AI · LLM · RAG · SQL
AI business assistant
The question is plain language. The answer has to come from the figures.
Ask a question. You see the calculation, the source and an action that waits for approval.
Demo on fictional data — never customer data.
This public demo calculates the answer from the dataset. No external language model is called. In production the same question would run through RAG on this model. A suggested action runs only after approval.
| Project | Margin | Target |
|---|---|---|
| PRJ-0142 | 0.0% | 28.0% |
| PRJ-0030 | 9.9% | 24.6% |
| PRJ-0021 | 12.8% | 23.2% |
| PRJ-0012 | 13.9% | 30.5% |
| PRJ-0087 | 14.8% | 30.0% |
Source: Noordlicht dataset, seed 20241008
SELECT project_id
FROM project
WHERE status = 'open'
AND (contract_value - actual_cost * 10000 / progress_bps)
/ contract_value < target_margin
ORDER BY 1;