An Experiential Approach to Enterprise Generative AI: A Case Study

11 min read

Listen to the latest articles and insights from our experts.

Listen to the latest articles and insights from our experts.

Summary

The Problem: A global educational publisher didn’t know where to begin with Generative AI (GenAI). Before it would invest in this new technology, it first needed to move beyond the industry hype and identify concrete use cases. 

The Solution: Tricon Infotech enabled the Client to experience GenAI first-hand by deploying a secure, private platform in the Client’s own environment and using its own proprietary content – all in the span of 60 days. This allowed the Client to quickly identify opportunities for GenAI to improve its business. 

The Process: 

  • Tricon built a “GenAI sandbox” that worked on the Client’s infrastructure, securely protecting its content and user activity. 
  • The new platform applied OpenAI’s GPT 3.5 and 4 to the Client’s proprietary content. Working together, the parties optimized the platform to perform a variety of useful tasks. 
  • At an in-person workshop, Tricon and the Client reviewed the system together and identified numerous ways it could automate time-consuming tasks. 

The Result: By experiencing GenAI first-hand in a live environment, the Client was able to identify prudent investment opportunities. The two companies now have a standing team of Tricon engineers who are actively building a solution that will enhance the Client’s workflow.

Generative AI: A Familiar Conundrum

Generative AI (GenAI) tools like OpenAI’s ChatGPT represent a paradigm shift in software products. For the first time in the history of computer science, a machine – when supplied with the proper context – can understand the intent of a user’s plain-language input. This is such a powerful tool that no one yet knows what this technology is capable of accomplishing. 

This situation has put many organizations in a position vis-a-vis GenAI similar to that of mobile applications when Apple first launched its App Store in 2008: Companies wondered if they needed a presence in the palms of users’ hands, what exactly they would offer, how they would develop these “apps”, and what the overall effect would be on their business. 

The Client, one of the world’s top educational publishers, felt a need to stay ahead of its competitors with GenAI, but before investing in the technology, it needed to answer two fundamental questions: 

  • What precisely should they do with GenAI? It was still unclear what value the technology offered and what risks it entailed. 
  • Who should undertake the necessary engineering work? The Client lacked in-house GenAI expertise, so it would be necessary to partner with an outside technology firm.  

The Client needed a way forward that would allow them to experiment with this potentially game-changing technology, but given the many unknowns, the substantial investment, and its business continuity requirements, it would need to proceed with caution. 

Fortunately, the Client already had a go-to, reliable technology partner that could help. 

A Familiar Partner for Uncharted Territory

The Client leveraged an existing relationship with custom software development company Tricon Infotech, which had helped the Client transition from a predominantly print-based strategy to a multimedia educational solutions provider. Over the years, Tricon had delivered a custom Learning Management System as well as specialized curriculum modules for PreK to postgraduate education. 

During the lengthy relationship, Tricon graduated from being a mere technical vendor to being a partner for strategy and ideation to envision, design, and construct new solutions. Given that Tricon was intimately familiar with the Client’s technical platforms and well-versed in rising GenAI technologies, it was an ideal partner for this new challenge. 

The Proposal

Tricon offered a solution that turned the usual engagement model on its head. The typical experience would involve a comprehensive discovery process, identifying the appropriate software solutions, and then building and implementing them. 

Instead, Tricon proposed creating a “GenAI sandbox” – a secure, internal prototype that the Client could use to experience GenAI technologies using its own content. This way, the Client could better identify challenges that the new technology could help solve.  

Furthermore, given the nature of the dynamic and evolving trends in GenAI, they agreed on a financial model that was built on mutual trust and shared risk: Tricon would deploy a team of engineers and specialists at its expense; if the initiative was successful and delivered the expected outcomes, then the two companies would agree to a commercial contract. 

Enter the Sandbox

Tricon Infotech is known for its unique teams-based model, in which engineers with a range of skills and expertise come together to work on a particular client’s needs. As the engagement expands, Tricon builds additional teams, each with a different focus, from new products to accessibility to Quality Assurance (QA) automation. 

For the GenAI sandbox, Tricon assembled a new team of over a dozen engineers with AI experience as well as some who were familiar with the Client’s architecture and technical environment. The group employed a “build in open” philosophy, with conversations taking place in real-time so that everyone was in sync and had complete context, an approach that encouraged a liberal sharing of ideas.  

The sandbox development proceeded quickly: 

  • The team set up a secure environment that would both protect and draw exclusively from the Client’s proprietary content, rather than third-party sources. 
  • The sandbox allowed connections to multiple LLMs, because different systems excel at different tasks.  
  • Anything that the GenAI sandbox created needed to be a) capable of being incorporated into the Client’s existing workflow, and b) reviewed by human Subject Matter Experts (SMEs) before being published or otherwise made public. 
  • Once the first version was set up and delivered to the client, the team continued to improve it by incorporating feedback from the Client’s SMEs, who would review and validate the system’s output. 

Tricon and the Client’s team decided to begin their GenAI exploration by incorporating into the sandbox one of their signature health sciences titles, a staple of medical school curricula. Tricon’s engineers configured the sandbox to import the publishers’ industry-standard EPUB files, which would make adding future titles easier. 

Over several weeks, the team created a chat-like dashboard that allowed the user to query the content and compare outputs from multiple GenAI systems (beginning with GPT 3.5 and 4) side-by-side. Prompts and responses were saved, allowing users to revisit past conversations using cognitive search capabilities. The engineers also participated in a Microsoft hackathon, which helped validate their methods and improve their prompt engineering. 

After 60 days, it was time to begin identifying what problems the Client could solve using GenAI. 

The Workshop

A group of employees from both companies gathered in person at a Client office. The group included not just technologists but also editors, SMEs, and specialists in products, sales, and marketing. They viewed GenAI through a wide lens, from automated content generation (creating new content) to automated content curation (adding value to existing content).  

After an initial review of the sandbox and an exploration of the vast possibilities presented by GenAI, the participants broke into three groups to brainstorm. Having experienced first-hand the power of the new technology, they were able to better envision a wide range of uses: 

  • Augmenting content. Once a book, article, or other piece of content existed in the new system, GenAI immediately could add value by writing captions or alt text for images, creating subtitles for videos, or even generating images and videos to illustrate text. 
  • Operational efficiency. Many tasks that involve comparing text versions – whether for QA purposes or for comparing publisher’s content with customer curriculum needs – could be partially fulfilled through LLM interpretation. Even if that only solved 70-80% of the requirement, the rest could be handled by human experts — a significant efficiency gain. 
  • New market opportunities. Educational materials could be converted into searchable reference materials that would be useful not only to teachers and students, but also to users in professional settings who need to query them for real-time, regularly updated answers. For example, a doctor working in a hospital could return to a version of a health sciences textbook as a familiar reference source, but now in a new, optimized format. 
  • Personalized content. GenAI could enhance existing text with summary explanations, test questions for students, or even complete machine translations into additional languages – all available for review by human SMEs before publication.  

When the participants reconvened after their breakout sessions, they shared over one hundred ideas. On the final day of the three-day workshop, they narrowed down the list to ten finalists, and then zeroed in on the four most promising ideas to pursue. 

Moving Forward

A guiding philosophy of the initiative was that GenAI won’t replace humans entirely, but it has the potential to make them more efficient, powerful, and able to focus on the tasks that truly require human expertise and judgment. Tasks that currently take one or two days often can be accomplished in one or two hours by using GenAI tools to analyze, compare, and create content.  

Beyond content creation, there are countless additional opportunities to employ GenAI within the educational publishing industry. For example, by feeding the entire corpus of the publisher’s content into the new GenAI platform, the publisher might map its content to the needs of a school or university’s specific curriculum, or use it to identify and fill gaps in a library’s holdings. 

At every stage, evaluating ROI is crucial. Each development involves an investment in engineering time, and each GenAI prompt involves a cost from a third-party provider. 

The Client and Tricon are now collaborating to turn the GenAI sandbox into a full-fledged internal product, with a standing team of Tricon engineers working exclusively on the Client’s AI initiatives. 

*Client name withheld for privacy reasons. 

Contact us for more information.

RELATED
Share Post: