Nextios, a company of LWSA, developed an AI-based Digital Teacher for Education – Technology and Innovation Ecosystem, a business area of Positivo Tecnologia. The solution was presented during Bett Educar 2024, an innovation and technology event for education held in April of this year. The Digital Teacher uses an architecture based on AWS services such as Amazon Bedrock, AWS Lambda, and Amazon Transcribe.
The Digital Teacher provides quick and effective responses, enhancing user experience. In addition to lesson summaries and test preparation, the solution has the ability to generate content automatically. The goal is to make the Digital Teacher widely available to schools, improving educational management and the learning experience.
Experimenting with AI as an educational tool
In addition to its product portfolio ranging from computers, smartphones to payment machines, Positivo Tecnologia has a business area focused on developing solutions for education, such as the Education Hub. Among the solutions in the Hub, the Aprimora platform last year received a generative artificial intelligence assistant – Maria – to help students understand specific issues and answer them.
According to Christiano Grillo Justus, head of Engineering and Architecture at Education, from then on, in 2023, conversations began with Nextios, an AWS partner, for the use of generative AI, leading to the project of the Digital Teacher, a solution integrated into the Education Hub.
The first concept was built in the Amazon Bedrock with content from the BNCC (Base Nacional Comum Curricular) and various teaching materials. “At this stage, she learned from these materials and, based on the user’s age, assumed what he should know by offering information in the voice tone of the schools associated with the system,” explains Justus.
According to Carlos Alberto Marangon, technical leader and AWS ambassador at Nextios, discussions with AWS led to the development of a new architecture model, previously unused. “The idea of using Amazon Bedrock’s Agents only in the final stage to bring knowledge base answers emerged, instead of developing the entire solution within the service,” he explains.
Teams created an architecture based on services like AWS Step Functions, AWS Lambda, Amazon OpenSearch Service, Amazon EventBridge, Amazon Transcribe, Amazon CloudFront, Application Load Balancer, AWS Certificate Manager and Amazon Simple Storage Service (Amazon S3). In Amazon Bedrock, they use the Anthropic Claude-3, Knowledge Bases and AI Agents.
According to Gabriel Leite da Silva, Partner Solutions Architect at AWS, the implementation strategy – isolated in specific tasks – allowed for a balance between generic information that should not be used, and very specific information that would make the solution similar to chatbots.
The specialist highlights the use of a technique called prompt chain, which combines low-code solutions with the code envelope that allows the final response to the user to be done through streaming, enabling the combination of business flow management with very fast responses from Amazon Bedrock.
“This combination of services is a paradigm shift. We opted for a serverless architecture and AWS tools that provide this low-code part, allowing us to not only deliver quickly but also giving Positivo Tecnologia the opportunity to experiment with different scenarios,” he says.
“It is not only focused on a knowledge base, being able to call a systemic API, understand its output, and perform systemic processing without the need for code to build these calls,” she says, also highlighting the possibility of integration with transactional and analytical workloads, with low code usage.
Meeting demands and performing tasks
Marangon, from Nextios, reinforces that the architecture, as designed, allows meeting demand and performing tasks for different levels of users, with access to specific content for teachers, students, and managers.
The use of Amazon Transcribe, for example, allows recorded classes to be saved and turned into summaries by Amazon Bedrock that can later be consulted by students. “The recorded class becomes searchable content for the student,” he explains, noting that the solution, with these components, made visible things that were not visible before and brought performance that was not possible before.
Justus states that the first proof of concept was developed in two months. Once the functionality was confirmed, a second version was prepared in 15 days, allowing the Digital Teacher to be presented at Bett Educar 2024. At this stage, integration with solutions developed by edtechs and offered in the Educational Hub was included. Thus, when consulted, the solution indicates which materials can be used within the existing solutions in the Hub.
“To present the Digital Teacher, we created a video of a class, showing that it can understand a video, allow students to ask questions, and teachers to create a more efficient study plan based on that specific class for certain students. In addition, the student could ask questions about the material and clarify doubts about the explanation, also based on the knowledge accumulated by AI,” explains.
Justus compares one of the functions of the Digital Teacher with that of a teaching assistant. “In the future, the principal of a school who wants to know how their classes are doing can view this information and receive management suggestions. It is an assistant for actions, feedback, data, and materials. That’s where we are heading,” he says.
The next step, according to Justus, is to present the solution to the business team. The goal of this process is to define the path to deliver the highest value in the shortest time, taking the solution to schools that can actually test them.
André Amorim, general director of Nextios, reinforces that “This project represents the first of many steps we are taking in building solutions with Generative Artificial Intelligence. The Digital Teacher is not just a project, it is a reflection of our commitment to staying at the forefront of technology and creating solutions using the most advanced tools for our clients,” he affirms.
