The enterprise knowledge hub that grounds AI agents in Roche-specific context — so they understand requests, cut hallucinations, and deliver outcomes that matter.
Advanced AI tools are only as effective as the context they are given. As these technologies become more complex, the greatest hurdle is feeding them the rich, specialized knowledge they need to actually understand the business.
AI tools have spread across every function — but they're trained on the public web, not Roche's data/metadata. They cannot reason over knowledge they have never seen.
Without a single source of Roche-specific meaning, every team tries to resolve the same problem — and AI keeps hallucinating, misinterpreting information, and returning results no one can rely on.
One governed hub that turns scattered Roche knowledge into context every AI agent can use — so answers are grounded, not guessed.
Speed to market, proven capabilities, and a clear operational model that lets domains focus on outcomes rather than infrastructure.
Enrich missing fields in metadata catalogs such as the R&D Dataset Portal (RDP), improving data completeness and downstream AI accuracy.
Maps synonyms — e.g. RO5541267 / atezolizumab / Tecentriq — so AI retrieves experimental data regardless of which identifier a researcher uses.
Links terminologies, ontologies, and master data across sources into one Knowledge Graph — so an agent can traverse from a compound to its targets, trials, and indications in a single query, instead of stitching together siloed systems.
Five guiding principles that shape how the Roche Semantic Hub is built, evolved, and measured.
Integrate Roche's terminologies, ontologies, reference and master data into one governed knowledge store.
Expose that knowledge through services that power AI and Therapeutic Discovery across the enterprise.
Build for genuine business needs — proven outcomes over theoretical completeness.
Continuously evaluate agent performance to drive improvement and demonstrate clear ROI.
Link the Hub to domain-specific systems for deeper, more accurate understanding of Roche workflows.
Three layers work together: data Sources feed the Roche Semantic Hub, which exposes knowledge via MCP and GraphQL API to a rich Consumption Layer of AI tools and users.
* Collibra integration is planned. For a technical overview of the hybrid on-prem / cloud architecture, see the appendix slides in the original deck.
To find out more about sources integrated into the Roche Semantic Hub and available services, visit Graph Content and Services.
Watch the narrated video walkthrough or reach out to the Roche Semantic Hub team to explore how the Roche Semantic Hub can reduce your time to production from 6 months to 2.
Watch Video Walkthrough