Document Processing
Novo Nordisk needed to shorten the slow and error-prone process for producing clinical study reports while keeping document generation deterministic and regulator-friendly in a tightly controlled pharmaceutical environment.
Reported impact
MongoDB helped Novo Nordisk cut report-writing times by 90%, reduce staffing required for CSR creation, and produce regulator-friendly drafts far faster while keeping humans in control.
Novo Nordisk built NovoScribe on MongoDB Atlas to support structured authoring for clinical study reports. Atlas stores approved text snippets, vectors, and case-specific trial data such as tables, lists, statistical outputs, and analytical results. The application uses a prompt-search-retrieval loop built around large language models and retrieval-augmented generation: Atlas retrieves relevant approved passages, the system presents five candidate text options to a human reviewer, and the chosen language is written back to the database before the next document section is assembled. That pattern keeps a human in the loop while still automating repetitive authoring work in a regulated setting where deterministic output matters. By combining reusable content modules with study-specific data inside Atlas, NovoScribe can build long CSR drafts section by section in minutes instead of weeks while reducing manual rework and keeping scientific staff focused on higher-value review work.
Approved text snippets and trial data are stored in MongoDB Atlas, the application retrieves relevant context for each prompt, a human reviewer selects from five candidate passages, and NovoScribe assembles report sections into a complete clinical study report.
"LLMs are probabilistic models, but pharmaceuticals [are] very tightly regulated, so we can't just assume that the model does the right thing."
"MongoDB has helped us cut writing times by 90%."
"There are plenty of other parts of the process that still need human thought and scientific input, so it's great to be able to pass the repetitive tasks on to computers."
"We need deterministic and clear outcomes, so we introduced an intermediate step where a human reviews five possible pieces of text and chooses the best one."
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
MongoDB Sample Dec 21, 2025 Novo Nordisk Novo Nordisk needed to reduce the 12-week, multi-team effort required to produce compliant clinical study reports without sacrificing quality or regulatory rigor. CSR creation reduced from 12 weeks to 10 minutes Amazon Bedrock · Claude 3 · LangChain 
Google Cloud Sample Archived · Oct 9, 2026 Angelus Founded by a dentist whose first product was based on a patent, Angelus competes on differentiation, developing bioceramic products and fiberglass post lines that require clinical proof, intellectual property… 50,000 pages of proprietary data now powering AI-generated documents Gemini MongoDB Sample Oct 8, 2026 Navisio AI Rapid adoption of Navisio AI's platform triggered severe latency spikes, elevated CPU and RAM consumption, and costly vertical scaling. As it prepares to scale to 1,000 by the end of the year, and eventually toward 5,000 users, the team is… MongoDB Atlas · MongoDB Atlas Vector Search 