Using Amazon Bedrock Agents to Accelerate Decisions Across Drug Development at AstraZeneca
AstraZeneca
AstraZeneca built Development Assistant, a multi-agent AI tool using Amazon Bedrock Agents, that lets clinical, regulatory, safety and quality teams ask natural language questions and get insights from structured and unstructured data in seconds via text-to-SQL generation and retrieval-augmented generation. A supervisor agent routes queries to specialized subagents (terminology, clinical, regulatory, database). Built on AstraZeneca's Drug Development Data Platform (3DP), the tool moved from concept to production in 6 months, including cybersecurity and AI governance checks, and is scaling to over 1,000 users.
Overview
AstraZeneca built Development Assistant, a multi-agent AI tool using Amazon Bedrock Agents, that lets clinical, regulatory, safety and quality teams ask natural language questions and get insights from structured and unstructured data in seconds via text-to-SQL generation and retrieval-augmented generation. A supervisor agent routes queries to specialized subagents (terminology, clinical, regulatory, database). Built on AstraZeneca's Drug Development Data Platform (3DP), the tool moved from concept to production in 6 months, including cybersecurity and AI governance checks, and is scaling to over 1,000 users.
The challenge
The global nature of AstraZeneca's clinical trial programs resulted in disparate systems, creating bottlenecks in analysis and decision-making; clinical teams spent hours manually compiling answers to questions like “Where are the highest-performing trial sites?” in an environment where speed to insight is essential.
The solution
AstraZeneca collaborated with AWS to build Development Assistant, a multi-agent AI tool using Amazon Bedrock Agents that lets clinical, regulatory, safety and quality teams ask natural language questions and get insights from structured and unstructured data in seconds via text-to-SQL generation combined with retrieval-augmented generation. A supervisor agent routes queries to specialized subagents, including a terminology agent, clinical agent, regulatory agent and database agent. The solution is built on AstraZeneca's Drug Development Data Platform (3DP), which ingests and aligns data from clinical, regulatory, quality and safety source systems with common vocabularies.
Reported business value
Development Assistant moved from concept to production in only 6 months, including cybersecurity and AI governance checks, and is scaling to over 1,000 users. Insights that once took hours are now available in minutes, and the system provides full transparency by showing which data tables were accessed and how results were generated.
Sources
Open any source and check the claim yourself — that is the point of the register.
Other pharmaceuticals entries in the register.
Novo Nordisk Scales to 2,500+ Use Cases with Secure Generative AI Using Amazon Bedrock
Novo Nordisk built a self-service generative AI platform on AWS, using Amazon Bedrock's foundation models, so employees could build and customize chatbots for nonregulated business use cases without needing to develop applications or maintain infrastructure themselves. The company worked with AWS Partner Cloud2 Oy (previously KeyCore) to validate the architecture for security and scalability. More than 25,000 Novo Nordisk employees have used the platform to create chatbots for over 2,500 unique use cases, such as retrieving information, drafting documents, or acting as a virtual colleague or critic. The company's general-purpose chatbot is used by more than 1,000 employees and processes over 26,000 prompts a month; its largest use case was trained on 140,000 documents. Each use case costs around $10 per month to run on AWS, using serverless services including Amazon DynamoDB and AWS Lambda. Building a chatbot now takes days rather than months, and some tasks that took a full day can be completed in as little as 10 minutes.
Novo Nordisk accelerates clinical insight with custom agents on Azure
Novo Nordisk wanted to accelerate pharmaceutical R&D decision-making by helping researchers explore and validate clinical hypotheses faster using AI-assisted quantitative analysis. Working with Microsoft's AI Acceleration Studio (Forward Deployed Engineering team), Novo Nordisk built a governed reasoning agent on Azure, drawing on its FounData initiative which harmonized more than 200,000 patient-years of clinical trial data aligned to CDISC, SDTM, and ADaM industry standards. The agent lets researchers ask complex scientific questions in natural language and generate and execute code against proprietary clinical datasets, with human-in-the-loop validation by biostatisticians before outputs influence decisions. Before production rollout, the teams ran thousands of automated tests and layered evaluation systems. The solution is reducing time to insight from weeks to minutes and increasing the number of scientific questions teams can evaluate from roughly 5-10 strong ideas per quarter to 50+.
Roche launches NVIDIA AI factory to accelerate the development of new therapeutics and diagnostics solutions
Roche deployed 2,176 NVIDIA Blackwell GPUs on premises across the United States and Europe, bringing its combined on-premise and cloud GPU infrastructure to more than 3,500 Blackwell GPUs. The AI factory is embedded across the value chain: NVIDIA BioNeMo powers Roche's Lab-in-the-Loop strategy connecting biological and chemistry experiments with AI models in R&D; NVIDIA Omniverse-powered digital twins optimize manufacturing production lines and factory designs; accelerated computing and NVIDIA Parabricks analyze diagnostics datasets and digital pathology images; and NVIDIA NeMo Guardrails support healthcare-grade conversational AI in digital health. The expansion builds on a NVIDIA collaboration that started in 2023, and Roche's U.S. subsidiary Genentech also leverages the infrastructure for drug discovery.
Aizon Improves Pharmaceutical Manufacturing With Artificial Intelligence Using AWS
Aizon built a GxP-compliant AI platform on AWS to help pharmaceutical manufacturers optimize biomanufacturing processes. Using AWS IoT, AWS Lambda, and Amazon SageMaker, Aizon created digital twins of bioreactors that trigger machine learning operations for process optimization. For a multinational pharmaceutical company producing blood-plasma-based drugs, Aizon built unsupervised and supervised learning models that identified two critical process variables; optimizing them improved yield by double digits, worth potentially tens of millions of dollars. AWS also reduced Aizon's compute costs by up to 80 percent.
Was this helpful?
Your feedback helps us improve our use case database