How Adisseo accelerated its regulatory compliance with generative AI
Adisseo accelerated its regulatory compliance (FDA/EFSA) by industrialising the conversion of its animal nutrition studies to the CDISC SEND standard in just 2.5 months, doubling the volume of studies processed while cutting lead times by a factor of 6.
Context
Converting 10+ years of heterogeneous studies (Excel, PDF, Word) from 21 external laboratories to the CDISC SEND 3.1.1 standard, with an incomplete ingredient reference dataset and no automated quality control.
Solution
Jinto deployed a Lakehouse pipeline built on Claude Sonnet 4.6 (via AWS Bedrock):
- Protocol-first approach: The study protocol is the source of truth.
- AI extraction: 8 parallel LLM blocks with strict validation (Pydantic / Instructor).
- Quality: 22 domain transformers and a blocking quality filter (Data Quality Gate).
- Deliverables: 22 CDISC SEND domains, 280,000+ published records and 3 bilingual (FR/EN) monitoring dashboards.
| Metric | Before (without LLM) | After (with Claude 4.6) |
|---|---|---|
| Processing time | 15 months | 2.5 months |
| Volume & sources | 30 studies / 1 lab | 70+ studies / 21 labs |
| Adding a study | 100% manual | YAML file (4 fields) |
| Data quality | Manual mapping | 100% match (1,887 automated tests) |