OttoTA | Devpost

Inspiration

The motivation behind OttoTA springs from a deeply personal place, echoing a common struggle among students: the daunting nature of textbooks causing many of us to completely ignore them, the scarcity of given practice problems and exams causing test anxiety and limiting understanding, and the limited availability and scalability of teaching assistants and professor face time.

These frustrations compound, and often leave us students without the support we need precisely when we need it. This shared challenge led to the conception of OttoTA, an AI-powered tutor designed to offer hyper-personalized educational assistance, capable of generating unlimited practice problems and providing answers to an endless array of questions.

Our resolve to create OttoTA was further strengthened by endless feedback from our peers, revealing a widespread desire for such a solution. In user interviews conducted while searching for a problem space and demoing an adjacent alpha-phase product that one of us built, we found collective yearning for an innovative educational tool underscored the urgency and relevance of our project. In fact, 87% of students stated that they believed that the provided practice material in most of their classes was insufficient.

Supporting our inspiration, scholarly research underscores the broader context of educational inequity that our project aims to address. Studies have shown persistent socioeconomic status (SES)-based gaps in academic performance from as early as kindergarten, indicating that disparities in educational outcomes are deeply ingrained and not solely confined to higher education. For instance, the Economic Policy Institute highlights that children from the highest and lowest fifths of the SES distribution exhibit significant performance gaps in both reading and math, which have remained largely unchanged over generations. This points to a systemic issue in educational access and the quality of resources available to students from different socioeconomic backgrounds (Economic Policy Institute).

Moreover, geography often dictates educational opportunity, with an "almost ironclad link between a child’s ZIP code and her chances of success," as highlighted by research from the Harvard Gazette. This geographic disparity mirrors achievement levels and demonstrates how societal inequities are entrenched within the educational system, further exacerbating the gap between privileged and underprivileged students (Harvard Gazette).

These insights from academic research and our own experiences converge to form the bedrock of our project's inspiration. OttoTA is not just a response to personal academic challenges but also a step toward addressing broader systemic issues in education. By leveraging AI to provide personalized, accessible educational support, OttoTA aspires to bridge the gap, offering a beacon of hope and a tool for empowerment in the face of daunting educational disparities.

We were also excited to join together as a team. Three of us (Matt, Tyler, Eashan) have gone to university together at Georgia Tech for three years, and the fourth member (Aadi) has close friends with Matt since freshman year of high school.

What it does

OttoTA is an hyperpersonalized AI-driven tutor designed to integrate seamlessly with a student's Canvas learning management system, revolutionizing the way students interact with their course material. By leveraging the power of artificial intelligence, OttoTA provides a personalized learning experience tailored to each student's unique needs and learning pace. Here's a breakdown of its core functionalities:

Integration with Canvas: By integrating directly with Canvas, OttoTA accesses relevant course materials and assignments to offer context-specific support. This allows for a seamless learning experience where OttoTA's assistance is directly aligned with the student's current coursework and learning objectives.
Dynamic Question-Answering: Students can ask OttoChat any course-related questions at any time. The AI uses NLP to understand the question, then uses a custom vector database built from their class materials, to precisely answer the question and provide detailed, step-by-step explanations, mimicking the interaction with a human tutor as closely as possible. This ensures students have 24/7 access to support, overcoming the limitations of traditional office hours.
Unlimited Practice Flashcards: Recognizing the importance of practice in mastering any subject, OttoTA generates an unlimited number of practice problems tailored to the pulled curriculum of each class, designed to challenge students appropriately based on their current level. Then, using the question-answering framework from above, OttoTA provides a solution to each generated practice problem. Each generated practice problem can be accessed by a student in the form of a flashcard, and flashcards can be continually generated.
Practice Exam Creation: To help students deal with test anxiety, OttoTA generates practice exams based on course materials, making sure questions are designed based on any practice exams and assignments. It also generates solutions using the dynamic question-answering framework.
Accessibility and Inclusivity: Designed with accessibility in mind, OttoTA aims to level the educational playing field by offering high-quality, personalized tutoring to all students, regardless of their socioeconomic background or geographic location. This inclusivity aligns with the broader goal of mitigating educational inequities.

By offering these features, OttoTA hopes to be a comprehensive educational tool that not only supports students academically but also empowers them to take control of their learning journey. Its innovative use of AI technology makes education more accessible, personalized, and effective, addressing the critical needs of today's diverse student population.

How we built it

We built a fullstack application including an LLM agent as part of the tech stack. We built our frontend using React and NextJS, and hosted the website on Vercel. Our backend is hosted on an Amazon EC2 instance of a FastAPI web server that we created. We used GCP SQL for our meta data database, GCP storage as a blob store, and Pinecone for vector storage for our RAG model. We used the Canvas API to access all the files in a given course to create the knowledge space for the AI to query through, and OpenAI’s GPT 4 model for question generation. Finally, we used Postman for creating a shared workspace allowing all teammembers to seamlessly test endpoints.

Challenges we ran into

Navigating the Canvas API proved to be challenging due to its unintuitive nature. Making it work seamlessly with our system required extensive testing and customization. Managing various file types that course materials can possibly be and student inputs also necessitated the development of robust processing capabilities to ensure compatibility and functionality without causing lots of errors. Finally, balancing the influx of data from different streams while minimizing latency was critical but super difficult. Ensuring real-time responsiveness while managing database interactions and AI computations presented a complex technical challenge.

Accomplishments that we're proud of

Successfully hosting our application across multiple cloud platforms, including Vercel, Amazon EC2, and GCP, demonstrated our ability to leverage cloud resources effectively, ensuring high availability and scalability. Integrating various data streams, from the Canvas API to our AI components, was a significant achievement. It allowed for a seamless flow of information, enhancing the functionality and responsiveness of OttoTA. Finally, achieving a fully cloud-hosted solution for OttoTA ensures that OttoTA is scalable, secure, and accessible from anywhere.

What we learned

Cloud Hosting: Navigating the complexities of cloud hosting across multiple platforms taught us valuable lessons in deployment, scalability, and security, broadening our understanding of cloud ecosystems. Delegating Work Efficiently: The project underscored the importance of efficient work delegation within our team. By playing to our strengths and dividing tasks strategically, we maximized our productivity and creativity. Full Stack Integration with LLMs: Integrating Large Language Models (LLMs) like GPT-4 into our tech stack was a learning curve. It offered us firsthand experience in harnessing the power of AI for educational purposes, from generating content to processing natural language inputs.

What's next for OttoTA

In the future, we are working to make OttoTA even more hyperpersonalized to each individual student. We are working on giving OttoTA the ability to analyze a student's grades and performance on practice exams to identify areas of strength and weakness, to then focus on topics where the student needs the most improvement, ensuring efficient and targeted learning. Additionally, we want to give OttoTA the capacity to continuously monitor a student's progress and improvement over time. Finally, we are building out features that would allow teachers to proactively interact with OttoTA, to offer it to their students and create group learning environments.