Regulatory AI workflows — the SPOR suite

A translation that took half an hour: mapping indications from a summary of product characteristics to the right MedDRA terms — preferred, low-level, high-level — sounds unspectacular. In my client’s regulatory routine, each procedure cost about 30 minutes of research, mapping and checking.

I built the AI skills that do it in minutes: extracting indications from the SmPC, resolving them to MedDRA terms, fully documented. The skills are in daily use; the team saves at least half to two-thirds of the time per procedure. The professional assessment stays with the regulatory team — the tools deliver the mapping and its source, not the decision.

A second building block: an automated compliance audit that validates the regulatory data fields in the client’s quality system against primary sources — with a traffic-light status and concrete correction proposals for every flagged field.

Both rest on an EMA SPOR integration (SMS, RMS, OMS) I delivered end-to-end: backend data loaders, matching algorithms, MCP tools, platform skills, and user documentation. And on a registry governing ~300 pharmaceutical data sources with license and risk analyses — so it is clear up front which source can legally be used for which purpose.

Built as a freelancer for symbiofy, implemented for a life-sciences company. The platform keeps evolving; I now work there as Solutions Consultant – Pharma.

For the technically interested

• MCP servers following the Model Context Protocol standard; platform skills access SPOR SMS/RMS/OMS, AMIce and EUR-Lex through these tools.
• Python/FastAPI backend; a token-count tie-breaker in the substance-search fuzzy matching prevents over-matches from descriptive suffixes (“Solution” for pure substances).
• EUR-Lex ingestion of the complete MDR/IVDR annexes, enabling granular questions on technical documentation, clinical evaluation and PMCF in chat.
• An 8-stage definition of done for data integrations — from infrastructure through MCP tools and platform registration to tenant seeding and documentation — as a binding quality bar.
• Section-wise caching of SmPCs for the field audit; knowledge tables as a revision-safe audit log.