A bilingual regulatory workspace for clinical instruments entering China

We built Dongchu's entire system, including an NMPA route and dossier constructor. A 2–5 person Algosoup team completed the engagement from June to July 2026, delivering in one month against an original expectation of three months.

The workspace begins by locking the proposed China intended use and product configuration. A qualified regulatory owner then records the applicable route and the evidence behind it. The fictional example shown here is a triple-quadrupole LC–MS/MS system intended for quantitative testing of human samples. The current Chinese classification catalogue points to an imported Class II medical-device registration route. That example cannot be generalised to every analyser or research-use configuration.

The route record also separates the foreign applicant, its appointed enterprise legal person in China, Dongchu's coordination role and any specialist reviewer. The software can retrieve relevant catalogue entries and expose ambiguity; it does not make the legal classification decision.

The control applies to translated or derived Chinese records; it does not require a word-for-word Chinese twin for every source file. Each controlled Chinese record points to the approved source, translator, reviewers, terminology and dossier destination. Repeated identifiers and claims, including model names, intended use, warnings and accessory scope, can then be checked across the package instead of being reconciled from memory.

The manufacturer remains the source owner. Dongchu and specialist reviewers can coordinate the China version without turning the workspace into a second, competing quality-management system.

The production workflow can use approved product language and a controlled glossary to prepare first drafts of dossier text, evidence requests and change-impact notes. Material statements should carry a source record and location. Missing or conflicting support should be surfaced for review and left unresolved.

Every output remains a draft until the appropriate bilingual, technical and regulatory reviewers record their decision. Model output must never set an approved, applicable, ready or submitted state. Imported documents remain untrusted evidence, and their contents stay isolated from system instructions.

The design maps the current official dossier structure and product-specific guidance to evidence records. Deterministic checks can identify missing signatures, stale versions, absent source links and inconsistent identifiers; AI may help extract and compare content. A regulatory owner still decides whether a requirement applies and whether the evidence is sufficient.

Each confirmed gap records what it blocks, who must resolve it and what evidence closes it. That creates useful operational visibility while the regulatory owner retains responsibility for determining dossier sufficiency.

The design maintains a versioned register of official rules, service pages, catalogues, guidance and standards. A source change opens an assessment for the regulatory owner to determine the affected products and records. Because official sites and formats can change, source monitoring needs a manual-review fallback and must never silently alter approved content.

Veeva, Rimsys and Ennov already market much of the broad feature set. The narrower opportunity is the cross-company China handoff: a product-scoped workspace for a foreign manufacturer, its appointed China agent, Dongchu, translators and specialist reviewers when occasional collaborators cannot practically work inside the manufacturer's enterprise RIM.

The production build focuses on the specific handoff left between those systems: source-to-Chinese traceability, NMPA-specific route and dossier views, controlled partner access, gap resolution and export back to the manufacturer's system of record. For organisations considering the same investment, an incumbent configuration or controlled service workflow may still be the better choice.

Before live evidence enters the system, the parties need to classify the data, map who can access it, including overseas support and model providers, and agree purpose, retention and deletion. The implementation baseline should include least-privilege access, encryption, audit logs, tenant separation, approved model endpoints and a commitment that customer data is not used to train public or shared models.

China's cross-border rules depend on the actual data and flow. The software can enforce an agreed boundary, but legal and data owners must determine which transfer mechanism or exemption applies; blanket claims that all material must stay in China, or may freely leave it, would both be unsafe.