关键指标
- 客户
- Caillau
- 行业
- 汽车工业
- 职能
- 质量部门
- 状态
- 验收中
- 部署
- 本地部署（容器）
背景
Automotive supplier working with many OEMs. First AI project in the group: systematically analyse customer specifications, standards and procedures.
问题
- Estimated €150,000 a year of manual analysis, with no guarantee of full coverage.
- Long, multilingual documents with nested requirements. A “drawer” omission is a non-compliance risk.
- Judgements were undocumented: facing an auditor, the answer was a memory, not a file.
方案
Seven-step pipeline: multi-format parsing, requirements extraction (including nested references), procedure indexing, semantic matching, scoring, justification comments, HITL validation before export.
调用的 NEXA 组件
| 组件 | 作用 |
|---|---|
| Arbitration algorithm | RAG + keywords + heuristics |
| Evidence Panel | Score, matched sources, lineage per requirement |
| HITL | Quality validation before export |
| AI Knowledge Vault | Reusable compliance patterns |
| Audit Pack | Standardised export for auditors |
影响
| 维度 | 具体影响 |
|---|---|
| Compliance | Systematic requirements analysis; fewer omissions. |
| Productivity | Reduction of the annual manual-analysis cost (est. €150,000). |
| Quality | Client score estimated 80/100 from the first demonstration. |
« 演示即给出约 80/100 的评分，有信心再经几次迭代收敛。 »