Roche

Roche’s patent organization had built several high-value GenAI tools for patent research, but attorneys faced a fragmented experience—switching between multiple interfaces, re-running queries, and manually rebuilding context to reach complete answers. Research methods were not orchestrated together, making it hard to scale workflows efficiently without duplication and added cost. As more AI tools were introduced, consistent governance for sensitive legal work became harder, increasing operational risk and dependence on external consultants.

Roche built Lex, an agentic AI interface in Dataiku that gives patent attorneys a single experience for research questions. Lex orchestrates RAG, full-text search, and deep search behind one interface, automatically selecting the right method per query and preserving context across follow-ups. Patent attorneys and paralegals defined real research needs while citizen developers extended agents in Dataiku; access controls, traceability, and C4 scoring/guardrails are embedded into workflows by design.

Roche avoided an estimated $375K–$475K in consultancy spend by enabling in-house delivery instead of relying on external consultants and IT. Underlying research services now support ~100 requests per week across dozens of active attorneys. Roughly 80 European patent attorneys and paralegals use the platform today, with plans to expand to more than 250 patent professionals globally.

• • Putting patent experts and citizen developers at the center of agent design reduces delivery friction and external consultancy dependence.
• • Orchestrating multiple research methods (RAG, full-text, deep search) behind one interface scales patent intelligence without fragmenting workflows.
• • Embedding governance (access controls, traceability, guardrails) at design time enables safer scale for sensitive legal research.