Blog

How Yseop Leveraged AWS to Develop a Unique Generative AI Application for Document Generation

December 5, 2024

Share This Post

Yseop has partnered with the world’s leading pharmaceutical companies to expedite the delivery of life-saving therapies and save thousands of hours in medical writing and review. Yseop Copilot, its enterprise platform, serves as a digital colleague, automating content creation for crucial regulatory documentation throughout drug development. The platform integrates seamlessly into medical writing tools and workflows and is already in use by several top 20 pharmaceutical companies.

As pioneers in Natural Language Processing (NLP) technology, Yseop leverages pre-trained LLM models specifically designed for the BioPharma industry. This multimodal approach marks a significant advancement for life sciences firms, enabling medical writers to boost productivity and focus on strategic tasks in a secure environment. Yseop Copilot utilizes proprietary prompts and validation methods to guarantee writing accuracy and traceability. Offering each customer a fully secure, private hosting environment.

Yseop and AWS: Empowering Innovation in Large Language Models

As a member of the AWS Partner Network (APN), Yseop has taken advantage of AWS’s robust security and scalability from the start. Now, Yseop is expanding its AWS usage to enhance its Large Language Model (LLM) ecosystem, infrastructure, and technical expertise.

By leveraging AWS, Yseop integrates LLM models to create a broader range of scientific content, all while meeting the strict security standards of the pharmaceutical industry. This collaboration empowers regulatory and medical writers at the world’s top pharmaceutical companies to significantly increase their productivity and accelerate the delivery of therapeutics to patients in need.

Key benefits of the Yseop and AWS partnership include:

Accelerating time-to-value with a ready-to-use SaaS enterprise platform powered by AWS.
Safely scaling content generation across organizations using AWS managed services and tools such as Amazon EKS, Amazon RDS, and Infrastructure as Code CDK.
Managing comprehensive data and document flows with enhanced security features.
Accessing cutting-edge AI capabilities with a model-agnostic SaaS supported by AWS SageMaker, AWS Inferentia, and AWS EC2.

AWS provides Yseop and its clients with a fully managed, private cloud environment, essential for the highly regulated life sciences industry.

Yseop’s Journey with AWS SageMaker

As the security and infrastructure needs of the life sciences sector evolved over the last five years, Yseop reached an inflection point regarding our hosting and infrastructure requirements. On the verge of a breakthrough in Generative AI for pharma, our team grappled with the limitations of our existing technology stack.

For years, Yseop relied on AWS SageMaker to deploy our custom AI models. This service provided full control of the AI stack, from training to testing and deployment, and supported us through numerous projects and implementations. However, starting last year, increasing project complexity and client demands in GenAI necessitated the use of multiple models for various use cases. As a result, we needed a more scalable, efficient, and cost-effective solution that allows fast evaluations of new frontier models and immediate productization.

When Bedrock reached General Availability (GA) in September 2023, Yseop began considering it a viable alternative. The specific models we needed became available in March 2024. With recommendations and support from our AWS contacts, Bedrock was chosen to address that need, and we transitioned smoothly within a few weeks.

Benefits of Bedrock for Yseop

Bedrock handles rapid deployments of pre-trained models, seamlessly integrating within our existing infrastructure. Most importantly, the application adapts to future needs, accelerating our pace of innovation. The move to Bedrock reflected Yseop’s strategic foresight and commitment to excellence, allowing the team to overcome previous scalability constraints and enhance technical ingenuity.

Bedrock is tailored to our unique life sciences requirements, launching models highly relevant to our industry. Clause 3.5, a competitor to GPT-4, stands out for its quality and sophistication. These pre-trained models are designed to scale, ensuring we meet customers’ demands without interruption.

While SageMaker remains part of Yseop’s toolkit for custom models, Bedrock’s rapid deployment capabilities better align with Yseop’s operational needs for developing applications securely. We can easily iterate on new models versions like Claude 3.5 and Llama 3.1 without needing extensive internal resources for updates and maintenance. These models are readily accessible and regularly upgraded, allowing our teams to focus on developing applications and leveraging the latest AI capabilities to remain competitive. Additionally, the Bedrock Custom Model Import feature facilitates the effortless integration of models built on Sagemaker.

Enhanced Capabilities with Bedrock

Bedrock brings a range of features that significantly enhance our operations, including comprehensive monitoring capabilities. Bedrock’s flexible pricing model, based on token consumption, ensures we only pay for what we use. This approach is especially beneficial for our inference tasks, where model utilization can vary significantly. The platform automatically adjusts resources based on the number of requests, ensuring optimal performance and efficient load management. This flexibility allows us to manage our budget more effectively while maintaining high performance.

Key features of Bedrock, such as Retrieval-Augmented Generation (RAG), agents, and fine-tuning capabilities, complement SageMaker’s strengths. RAG allows for seamless content retrieval through advanced integration with NoSQL databases, vector databases and content storage systems (Knowledge Bases), providing more accurate and relevant responses, enhancing the overall user experience.

Bedrock integrates smoothly with other AWS data services, aligning perfectly with our existing AWS-based architecture and simplifying our daily management. The consistency of having a single point of entry, regardless of the model, streamlines our operations, allowing our team to adapt quickly and efficiently to different models.

Complementary Solutions: Bedrock and SageMaker

We see Bedrock and SageMaker as complementary solutions. Bedrock offers easy access to a variety of pre-trained models, saving time and accelerating development, while SageMaker provides full control over our custom models. The transition to Bedrock was about using the right tool for the right task. For example, while we anticipate the need for future grammar constraints to ensure consistent and high-quality responses, this feature is not yet available on Bedrock. Therefore, we continue to utilize SageMaker for specific applications where consistent response formatting is crucial.

By adopting Bedrock, Yseop has significantly improved efficiency and productivity, allowing our LLM team to concentrate on high-impact tasks that drive business growth. The ability to quickly deploy and manage pre-trained models without extensive internal resources has been transformative, enabling us to remain competitive in a rapidly evolving industry. The combined use of Bedrock and SageMaker creates a versatile and robust AI ecosystem. This strategic approach ensures that we meet a wide range of AI requirements, delivering superior value to our clients.

Conclusion

Yseop’s transition to Bedrock, AWS’s latest AI and machine learning platform, has been transformative for our team. This strategic move, following intense discussions and rigorous evaluations, demonstrates our foresight and unwavering commitment to excellence. Yseop and AWS align in their vision for advancing Generative AI in the life sciences sector. The Bedrock platform allows us to overcome scalability limitations and advance our technical capabilities.

Our existing partnership with AWS played a significant role in this transition. With AWS’s extensive support, integrating Bedrock into our operations was effortless. This integration streamlines our model deployment and management processes, enabling us to focus more on our core strength—innovation.