CACTUS Communications builds highly scalable generative AI-powered services on Amazon Bedrock

With offices in Princeton, London, Singapore, Beijing, Shanghai, Seoul, Tokyo, and Mumbai, CACTUS supports researchers, publishers, universities, and life science organizations through a range of human-driven editorial and research support services, as well as AI-powered tools. In 2019, it launched a dedicated research and development division, CACTUS Labs, to drive the development of AI tools for services like automated grammar correction. In 2021, the division began developing an LLM for summarizing texts and generating content. 

 

However, with companies like Anthropic spending billions of dollars developing general-purpose LLMs, CACTUS recognized it could gain more by building on those advancements. Parth Agrawal, associate director of ML and AI at CACTUS, explains, “With LLMs advancing worldwide, we recognized a clear opportunity to concentrate on building generative AI services that meaningfully enhanced our portfolio.” 

 

CACTUS wanted to work with the latest LLMs without having to provision any underlying infrastructure. “We found that running training and inference clusters was costly because they required significant overprovisioning to ensure stability,” says Agrawal. “We didn’t want to dedicate resources to infrastructure management when they could be better used on innovation.” The company sought an enterprise-ready solution that provided secure access to LLMs and scaled automatically. 

CACTUS opted for Amazon Bedrock, a fully managed service that provides access to a selection of high-performing foundation models (FMs) from leading AI companies including Anthropic. Using Amazon Bedrock, CACTUS can innovate at speed—building and scaling FMs through a single API - all while maintaining a high level of data security. “We work with unpublished research, so safeguarding it is crucial,” says Agrawal. “After years of running mission-critical workloads on AWS, we have complete confidence in their ability to protect our most valuable assets.” 

CACTUS deployed Amazon Bedrock into an existing landscape of modernized AWS services, including Amazon Elastic Kubernetes Service (Amazon EKS) to manage Kubernetes clusters, AWS Lambda for serverless processing, and Amazon Athena for analysis of data stored in Amazon Simple Storage Service (Amazon S3) buckets. To simplify development with Amazon Bedrock, the team used Terraform and AWS Cloud Development Kit (AWS CDK) to automate infrastructure provisioning. CACTUS also integrated Amazon OpenSearch Service to store and index data for LLMs to process.

The company found that Amazon Bedrock running Anthropic Claude delivered the best price performance. It also noted that Amazon Bedrock Guardrails, which provide configurable safeguards for building generative AI applications, helped block harmful or biased outputs and maintain privacy standards. “With Amazon Bedrock Guardrails, we didn’t have to break our heads on building a robust security system,” recounts Agrawal. 

Using Amazon Bedrock, CACTUS began working on around 20 generative AI use cases for its services. These included Ask R Discovery, a research assistant available within CACTUS’s academic discovery platform R Discovery, and a suite of AI tools for accessibility such as Alt Text to create image descriptions. It also developed ChatPDF, a document-interaction application that is integrated into Paperpal, the company’s AI-powered writing toolkit. Ravish Tiwari, senior architect at CACTUS, says, “ChatPDF went viral overnight, and it became Paperpal’s hero feature, accounting for 9–11 percent of all generative AI prompts. With the feature’s serverless architecture and Amazon Bedrock scaling automatically, performance remained reliable as workloads increased.” To keep costs under control as usage grew, the CACTUS team worked with AWS to fine-tune its generative AI architecture. “By using prompt caching on the advice of the AWS team,” says Tiwari, “we reduced our token usage costs by 5–10 percent, making usage growth sustainable.”

Paperpal has become the fastest-growing product in the history of CACTUS’s portfolio, supported by its AWS-based generative AI capabilities. By mid-2025, it had more than 3 million registered users across 125 countries, with a 40 percent increase in daily active users engaging with its generative AI tools. “Amazon Bedrock gives us the scalability to meet the huge demand for generative AI-powered services among researchers and academics worldwide,” says Agrawal.

By combining Amazon Bedrock with its existing cloud-native AWS infrastructure, CACTUS has optimized the speed of generative AI development. Says Agrawal, “We’ve reduced idea-to-prototype cycles for new AI capabilities by 65 percent, and that’s a direct result of using generative AI on AWS. In addition, our core generative AI-driven products keep a healthy net promoter score range of around 50–70 against an industry average of 20−30 for non-academia-centric products,” comments Agrawal.

For the past 3 years, CACTUS has experienced exponential growth of over 100 percent annually in its AI products and solutions portfolio—a level of expansion comparable to that achieved in 10 years when services were solely human driven. Agrawal concludes, “We’re in a great position to continue our present rate of growth by leaning into our 20-plus years of industry experience and cutting-edge AWS generative AI technology. We have the credibility and trust among researchers and academics, and now we have the technology to transform what’s possible.”