SoftwareOne case study

GenAI helps Worldreader build children’s reading skills

An international nonprofit, Worldreader wanted to explore the use of GenAI to more quickly and easily create digital activities that encourage child readers to read more and build comprehension.

Founded in 2010 and headquartered in San Francisco, California, Worldreader has a mission to encourage children around the world to read at least 25 books a year with understanding. Its free app, BookSmart, provides children and adults with access to digital books, reading challenges and comprehension-building exercises.

Since it began operations, Worldreader has helped 22 million readers across 100 countries to read 77 million books. In 2023, it received the US Library of Congress Literacy Awards’ International Prize, which recognises exceptional efforts to increase literacy around the world.

Scalable solution
that can quickly generate content for as many books as needed
Pay-as-you-go
solution that’s cost-effective for Worldreader
Fine-tuned outcomes
that tailor content for children with a wide range of backgrounds

Client: Worldreader
Industry: Nonprofit
Platform: AWS Cloud
Services: Cloud Services: Data & AI Services: Amazon Web Services: Amazon S3, Amazon Bedrock, AWS Lambda, AWS Step Functions, Anthropic Claude 3 LLM
Country: United States

Exploring GenAI as a way to improve reader engagement

As an international nonprofit organisation with fewer than 100 staff members, Worldreader is always looking to work more efficiently so that it can focus on its mission: encouraging children around the world to read at least 25 books a year with understanding. Children who are more engaged and regular readers, it believes, are more likely to become better educated and more emotionally intelligent adults with a higher lifetime potential for earning.

Worldreader wanted to explore technologies that could help children learn and retain more from the digital books they read, and to make those books and related reading materials available in children’s native languages. After receiving grant funding from Amazon Web Services (AWS), its cloud service provider, Worldreader turned to AWS Premier Partner SoftwareOne for help.

The project launched with an ambitious question: Could Worldreader improve engagement and learning in a resource-efficient way by using advanced cloud services like artificial intelligence (AI), generative AI (GenAI) and machine learning (ML)? Recognising that SoftwareOne had the expertise and experience to deliver such solutions, and supported in part with funding from both AWS and SoftwareOne, Worldreader set out to discover the answer.

Creating reader activities is a big job for a small team

SoftwareOne started with a focus on Worldreader’s BookSmart, a free app that children – and the adults who support their reading goals – can use on any device to read digital books, compete in reading challenges and participate in comprehension-building activities.

Worldreader wanted to see how technology could make it easier for the organisation to create and personalise those comprehension activities. “We have a very lean team,” says Sonny Lacey, Worldreader’s director of product. “We have a wealth of books and a wealth of ideas, but the creation of activities – meaningful activities that are specifically aligned with a book – that’s a labour-intensive effort. We had been thinking, ‘What are some ways we can leverage GenAI to be an extra hand in the factory here?’”

SoftwareOne invested in this project by providing professional services through its Impact program which focuses on creating positive impacts in sustainability and community engagement. SoftwareOne built a solution for Worldreader using machine learning and GenAI services that fosters a culture of continuous learning by creating new comprehension activities based on Worldreader’s existing collection of reading materials. Worldreader uses the solution to drop a digital book into an Amazon Simple Storage Service (Amazon S3) bucket. AWS Lambda compute service triggers code to generate content using Amazon Bedrock and Anthropic’s Claude 3 Opus large language model (LLM), producing interactive exercises that can be personalized for a wide range of readers.

Using the Amazon Bedrock fully managed service for building and scaling GenAI applications, SoftwareOne created prompts on Claude 3 Opus. The approach established a way to provide the full text of any Worldreader book in local languages and then create rules for generating relevant and appropriate book-aligned activities to engage readers.

These rules helped to fine-tune how activities take into account variations in children’s socioeconomic status and which adults they are exploring reading activities with. This ensures that activities are appropriate for children from a wide range of backgrounds, whether they are reading with a parent or with some other responsible adult. The LLM can also analyse books in other languages like Swahili, enabling Worldreader to generate activities for a near-native speaker experience.

The solution allows Worldreader to take a digital book and complete a language translation that previously took days in less than one minute. This enables the creation of appropriate book-aligned activities in its application to provide a personalised experience for a wide range of readers.

Everything was just spot on and creative.

Sabrina Abreu

Content Acquisition Manager, Worldreader

AWS and SoftwareOne
Get AWS advisory, migration, and managed services from an AWS Premier Tier Services Partner.
Learn more

Scalable and efficient, with a fresh perspective

The solution is easily scalable and designed to produce new reader activities for one book at a time or for many at once, meeting Worldreader’s need for efficiency. As a nonprofit, controlling costs is essential and if the app doesn’t need to generate activities for a book, Worldreader doesn’t have to pay for services it’s not using. But in periods of high demand, the solution can create new content for 1,000 books at once.

Because the solution is based only on Worldreader’s own data on AWS, it’s also secure, adding peace of mind. Amazon Bedrock enables users to experiment with secure foundation models and privately customise them with their own data, without having to use their intellectual property to train a public model.

Sabrina Abreu, content acquisition manager for Worldreader, says she’s been impressed by the quality of the activity materials that the solution has been able to generate.

“It just went beyond my expectations in terms of the output,” she says. “In many cases, yes, we had to refine a little, but that happens with humans too. Everything was just spot on and creative. Once you’ve done this for a few years, you see the same activities crop up. There are new activities, things that we’ve never thought of doing that way, which was really great. For the end user, that’s awesome because you’re not seeing the same recurring activities. You’re reading a new book, and this has a new perspective or something new to look at.”

The solution now allows Worldreader to take a digital book and complete a language translation that previously took days into less than one minute enabling creation of appropriate book-aligned activities in their application that provide a personalised experience for a wide range of readers.

Looking ahead, Worldreader is considering ways to further personalise activities – for example, by tailoring different activities for children who live in remote rural areas of Kenya versus urban neighbourhoods in Nairobi, or for incarcerated adults who share reading activities remotely with their children.

Worldreader is also thinking about other ways in which it might use GenAI to do things like assess a child’s reading comprehension using generated quizzes or games.

“I just also want to add: it was fun,” says Lacey. “So much of work can be drudgery. But it was all fun, and that’s a great thing.”