Accelerating article research using Amazon Nova Lite with EBSCO

Researchers and students frequently face the daunting task of finding relevant scholarly articles to inform their projects. For EBSCO’s customers, this meant searching the company’s growing library of over 57,000 journals, then reading dense, unstructured abstracts or skimming lengthy full-text articles. “Finding the right resource can be like finding a needle in a haystack,” says Michael Napoleone, vice president of product management at EBSCO.

EBSCO saw an opportunity to use generative AI to help users quickly and effectively evaluate articles’ relevance to the research in hand. The company envisioned a feature that would summarize each article’s subject matter and findings in a few succinct points. The feature aimed to maintain the accuracy, relevance, grounding, and intention of the research knowledge without replacing the value of the full text. EBSCO reached out to the AWS team to further develop the concept, which led to an engagement with the AWS Generative AI Innovation Center to build a working prototype. This deep collaboration would lay the foundation for an innovative and responsible solution for EBSCO’s customers.

EBSCO created the AI Insights feature by using Amazon Bedrock—a comprehensive, secure, and flexible service for building generative AI applications and agents. The company had initially used a third-party foundation model but saw room for improvement in latency and user experience. The team decided to switch to Amazon Nova Lite, a multimodal model that processes image, video, and text with high speed. As a result, EBSCO could lower latency and maintain high-quality results, driving a seamlessly enhanced user experience.

The solution works as follows: A user chooses the AI Insights button on an article. The solution fetches the document’s full text at that moment, injects it into a prompt, and sends the prompt to Amazon Bedrock. Then, within seconds, Amazon Nova Lite analyzes the content and generates a summary with two to five bulleted insights.

Since EBSCO offers academic articles in 33 core languages, the fact that Amazon Nova Lite offers support for over 200 languages was another significant benefit. Amazon Nova Lite generates an English summary based on an article in any language, then automatically translates based on the selected language of the user interface—all without an extra step in its digital architecture. Domain experts rigorously evaluate the model so that the results maintain EBSCO’s standards for delivering high-quality consequential knowledge, even when processing sensitive academic subjects. The high speed and low latency of Amazon Nova Lite also help EBSCO significantly reduce its data storage requirements because the AI Insights feature runs in memory at runtime.

AI Insights surpassed 120,000 summaries generated 6 months after its launch. Users spend more time using EBSCO’s products and are more likely to read the full text of the articles that are surfaced, citing the feature as a significant time-saver. “Now, using Amazon Nova Lite, you choose the AI Insights button, and a response directly comes streaming in,” says Annette Krawietz, principal product manager at EBSCO. “It was a huge improvement.” By switching to Amazon Nova Lite, EBSCO improved latency by 30–40 percent and reduced response times from 10 to under 5 seconds. The company also cut inference costs by 98 percent compared to the previous model.

With the success of AI Insights, EBSCO paved the way for further innovations. It recently released Natural Language Search, a feature that helps users find the academic content they need through naturally worded phrases. Since launch, the feature has processed over 4 million searches. The company also plans to launch Recommended Searches, a feature that suggests alternative searches to help users dig deeper into their topics.

EBSCO will continue working alongside an extensive network of AI specialists from AWS to drive innovation through the use of responsible AI in the library services industry. “We’ve gotten a tremendous amount out of our work with the AWS team,” says Elliot Rodriguez, senior systems architect at EBSCO. “I’m eager to see the road map for where we can take our AI features going forward.”