
STAYON : From Scattered Information to Connected Action
Students have to process information from numerous scattered sources: college circulars, assignment PDFs, internship and scholarship alerts, hackathon details, emails, notes, and personal goals. STAYON is an AI-powered student execution assistant that unifies one's fragmented learning resources, goals, tasks, schedules, habits, and deadlines into a collective workflow that turns one's information and intentions into organized, executable plans of action.
The Problem We Wanted to Solve
Students get important information from a variety of disparate sources: college circulars, assignment PDFs, hackathon details, emails, personal notes, academic or even personal goals.
The problem is not in finding this information. It's in acting upon it and following through.
A deadline could be hidden at the end of a long document. There might be eligibility criteria for an opportunity that you're not aware of. You could have a huge personal or academic goal in front of you and have no idea how or where to start working towards it. Students also use multiple apps for different tasks, for reminders, for planning, for habit tracking, for notes and more.
We wanted to bring this disparate information together into a connected and easily accessible experience that lets people not just consume the information as static data, but work with it in concert to accomplish something, to follow through.
Introducing STAYON
Stayon is an AI assistant for students that ties together disparate learning resources, goals, tasks, schedules, habits, and deadlines in one connected workflow.
By connecting learning resources, goals, tasks, and schedules into a single system, STAYON enables you to take your scattered notes and ideas and build on them step by step, from initial understanding to planning to execution.
You can begin with a document, or a goal, or a task, and then layer additional context and requirements on top until you've built a comprehensive plan that can be acted upon, monitored, and refined.
Students should be able to take their research and planning process, which currently exists in many disconnected pieces, and form it into a connected whole that moves from an understanding of information to planning to action.
From Information to Action
One of the main features of STAYON is that it is based on the information a student may already have. Instead of forcing students to extract every valuable piece of information from a document individually, STAYON helps organize the data to fit their planning purposes.
Right after a document is uploaded, it is analyzed, and valuable pieces of information are extracted and organized as title, deadline, eligibility, requirements, important information, and action items. Thus, a student gets a precise understanding of the information they have.
However, the user should not forget that they are the ones who have overall control over the software. Based on the information extracted from the documents, the student can define a goal and choose what they want to achieve. STAYON can help prioritize actionable items and arrange them as a planning scheme.
What STAYON AI Brings Together
STAYON is organized around a set of interconnected resources and activities that can be accessed from a single workspace. The navigation is based on Knowledge, Goals, Tasks, Schedule, and StayOn AI, with the Home dashboard aggregating the most relevant information in one place.
Knowledge : It offers a dedicated space to store and manage learning or resource documents, with STAYON extracting relevant information from the uploaded PDF files and presenting deadlines, requirements, criteria, and other action items for easy planning and management.
Goals : It is created to represent the desired outcome, with the user being able to define a title, description, deadline, and attached documents for each of their goals, providing a clear objective to be accomplished.
Tasks : It is used to break down the itemized actions needed to achieve goals, with the ability to organize them as per their due date and deadline, define the time required and priority level, and attach them to a specific goal, and view active tasks, including those due today, overdue, and completed.
Schedule : It is utilized to assign specific tasks to time slots, as opposed to having a linear list of activities, transforming the task list into a full planning and time management tool.
StayOn AI : It is a natural language interface that allows the user to interact with their learning and planning data, asking for task details, getting assistance with goal breakdown, studying, and editing their plan.
Home : It acts as the central dashboard, allowing the user to have an overview of the current learning, goals, tasks, activities, and the ability to access the StayOn AI interface.
Adaptive Replanning
Planning does not always go as expected. A student can forget to do one task, get distracted with another activity, or a new responsibility arises. STAYON can adapt to new circumstances instead of having the student stuck to a schedule that is no longer relevant.
By having the unfinished work, the app can propose a new plan depending on the tasks that were not completed.
The suggested modifications are displayed to the user along with the changes that will be executed on their planning. They can then validate the modifications and commit to the new schedule.
This way, the process of planning, execution, advancement, and replanning is continuous, and not stuck to one single plan.
Building STAYON with AWS
STAYON is built as a cloud-based application using AWS services in its authentication, API management, compute, storage, database, and AI capabilities.
For authentication, Amazon Cognito was utilized to manage users and their credentials. Upon successful authentication, the application uses the ID token to gain access to its APIs.
The application’s API layer was built using Amazon API Gateway, whereas Lambda functions were used for its business logic, including documents, goals, tasks, daily view, AI agent, and replanning. For storage, the application utilizes Amazon DynamoDB to store its data, including users, documents, goals, tasks, habits, and habit logs. In addition, the application stores PDF and document files in Amazon S3.
Finally, the application’s AI layer uses the Gemini API to support document understanding, goal and task generation, conversational assistance, and replanning features. The application is also designed with a provider abstraction layer, allowing the AI provider to be replaced or extended in the future without changing the core application workflow.
System Architecture
The general architecture of this design is organized in a specific flow from the student’s interface to the backend and artificial intelligence (AI) layers.
The student’s side of the platform is built with React, Vite, and TypeScript, while their requests to the API are managed with Amazon Cognito upon which the authentication system is built.
Authenticating the user allows the requests to the application programming interface (API) via Amazon API Gateway to be fulfilled by AWS Lambda, where the application’s business logic runs.
The service further communicates with Amazon DynamoDB to manage the application’s relational database and Amazon S3 to store the documents uploaded by users.
Finally, the AI layer connects the service to Amazon Bedrock and Amazon Nova 2 Lite to perform document understanding, goal and task generation, conversation, and replanning tasks.
This architecture is segmented into a set of components that together realize the functionality of the STAYON service as a unified whole while being built on separate technologies for the frontend, backend, database, and API for managing the interactions between the students and the service.
Testing and Engineering
In addition to designing the connection of services and developing the interface, a critical aspect of the work was the evaluation of each segment of the application during development.
The frontend TypeScript type check and prod build were successful. On the backend, our test suite recorded 103 passed tests and 0 failures, while the AI test suite had 86 passed and 1 skipped. We also did a live test with the Gemini provider, which had 3 tests passed.
Cognito token refresh had a race condition problem, where many requests could come in before the token is refreshed. We solved it by introducing a singleton refresh promise that will be shared between all refresh requests.
This way, tests and fixes helped us not only to validate the individual features but also to ensure the application’s stability as the different parts are put together.
What Makes STAYON Different
STAYON is a concept, which is based on the idea of connecting resources and activities, in opposite to the isolated problem-solving for one part.
A specific document may contain the information for the user, a goal is an objective that the student wants to accomplish, which in turn requires certain tasks, a schedule brings these items together with time-planning, and the AI assistant helps perform and optimize these steps with the person’s cooperation — these elements form the system working in accordance with the user instead of isolating functions.
Another essential notion is that the student using the service remains in charge. STAYON can recognize the data, offer objectives, suggest tasks, propose schedules, and optimize the plan, but the user must approve all alterations and authorized actions.
Therefore, the information is arranged in a cycle from data input to the goals, tasks, and schedule, followed by the execution and further analysis of progress and adjustments. Thus, all parts are connected as a unity for the user in contrast to the traditional planning methods.
What We Learned
Developing STAYON revealed to us the importance of correctly designing an AI-driven application. We came to the conclusion that the key to success is the effective combination of the language model’s power and the application’s tools.
We had to consider all processes where information had to be passed from one part of the system to another. For instance, we had to decide how to train our model using document parsing, data structure, and goal and task trees. Moreover, we needed to establish the necessary connections between such processes as agenda creation, tracking, scheduling, and AI-driven replanning.
Yet, another valuable experience was gained when we learned how to implement AI into the application correctly. In particular, we needed to ensure that the AI model could make changes directly in the application’s database. However, we had to find a way to prevent destructive operations and keep the user in charge of all ongoing processes. Nevertheless, our model demonstrated that it was possible to effectively implement such interactions.
Finally, we have managed to build an application compatible with AWS services. This allowed us to create an effective cloud system where user authentication, API services, serverless computing, storage, databases, and AI services are integrated.
Conclusion
The STAYON AI was conceptualized with a rather straightforward idea in mind: students have plenty of readily available information and resources at their disposal, but it is all fragmented and disconnected from the execution process;
By unifying the aspects of Knowledge, Goals, Tasks, Schedule, and StayOn AI itself into a single cohesive unit and, thus, establishing an ecosystem where information and execution are interoperable, the product at hand effectively addresses the identified issue. The tool is designed to facilitate multi-faceted information management, goal setting, task scheduling, and AI utilization with the aid of advanced but accessible scheduling and replanning technologies and the user’s active involvement in the decision-making process.
The software is further supported by a vast array of services provided by AWS, including Cognito, API Gateway, Lambda, DynamoDB, S3, and Bedrock. Overall, the STAYON AI is a prime example of how cutting-edge technologies can be utilized to design a truly compelling product that synergistically integrates disparate pieces of information and action into a unified whole.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article