Amplifire Case Study

ChatGPT Integrations Enable E-Learning Company to Automate Content Authoring

The Opportunity Develop a Proof of Concept to Use Generative AI to Automate Content Authoring

Amplifire is an e-learning platform that utilizes cognitive science principles to help individuals and organizations improve their knowledge retention and critical thinking skills. The platform employs a unique approach that involves targeted questioning and feedback loops to optimize learning and retention. Amplifire’s AI-powered algorithm adapts to each user’s performance, delivering personalized experiences designed to enhance learning outcomes, boost confidence, and improve decision-making abilities across various industries and disciplines.

Aside from their learning platform, Amplifire offers instructional development services to their clients who need help creating training modules. Authors often must ingest copious amounts of data and material to develop course curricula and assessments. With the popularity of ChatGPT and other AI for content generation, Amplifire sought to use these tools as a valuable way to streamline course development processes. The newly released OpenAI API is now available for business applications, making it an attractive option for Amplifire to leverage.

Interactive prototype of ChatGPT integration

Our Response Mastering the Art of ChatGPT Prompts and Creating Boundaries to Limit the Scope of AI

MojoTech was tasked with conducting a concept validation study to determine the suitability of ChatGPT for Amplifire’s use case. ChatGPT and other generative AI can be very helpful when generating unique content with vague or philosophical questions. However, Mojotech wanted to ensure highly relevant content by restricting the scope of what is generated to a specific document. To achieve this, MojoTech created a simple proof of concept to test GPT 3.5 and determine whether AI could generate quality text.

Document and Data Ingestion

One of the challenges that MojoTech faced was determining whether the API could handle document and data ingestion. Amplifire’s Authors and subject matter experts often use 50 pages or more of source material documents to create course content. It was essential to test whether the API could import PDFs. MojoTech used a customized library to extract text from PDFs and feed it to the API. However, there were limitations to the amount of text that could be processed at once, with the API maxing out at 3000 words (tokens). To overcome this, MojoTech sliced the document into smaller chunks and fed them to the API.

Mastering ChatGPT Prompts

MojoTech placed significant emphasis on the input prompts given to ChatGPT as it impacts the quality of the generated output. To ensure that the results returned were high quality, MojoTech wrote the code to collect all the relevant chunks of information and combine them to create the holistic scope of context needed for outbound prompts. The prompts were carefully crafted to generate accurate summaries, concise outlines, and multiple-choice quizzes.

Interactive Prototype and Front End

MojoTech created a custom front end for ChatGPT that helped control any variability in prompting the AI, which could adversely affect its output. It also simplified user interaction, allowing file uploads and a single-button operation. We used a newer technology stack to ensure flexibility, allowing fast iteration and low-code writing. It enabled the team to quickly deploy a functional website with login, document upload, and content generation features.

The Outcome Prototype Generates Effective Content and Helps Reduce Authors’ Workload

During the feasibility study, MojoTech provided daily progress updates to Amplifire and prioritized the next features for the proof of concept. Over a two-week sprint, MojoTech successfully delivered a working prototype that can generate a summary, outline, and assessment questions from a massive PDF document in a fraction of the time it would take Amplifire’s content authors to create the same content.

While the prototype is not meant to supplant authors, it significantly reduces the workload for authors and subject matter experts who must create the appropriate content. To test the prototype’s effectiveness, we gathered feedback from actual users, including CFOs and professionals in the financial sector, to determine if the generated summary of content relating to an auditing document was enticing and valuable to them. We garnered resounding positive feedback from both the users and the client.

You also might be interested in...

United Way’s modernized web and mobile app.

Modernization

United Way: Web App Modernization Helps Users Find Essential Services Easier

Smart search, database design, API development & mapping UI unlock new ways of connecting users to services.

View Case Study

Healthcare

Docent Health: Product Discovery Leverages Tech to Build Patient Relationships

Solution prioritization & prototyping make way for engagement tools focused on improving patient experiences.

View Case Study