Friendly Awesome Smart Tutor

Insufficiency and absence of a varied array of resources in early-life education remain the major indicator of long-term socio-economic status, professional prosperity, and health or happiness indicators. The Commonwealth Institute has demonstrated a cyclic mechanism reinforcing systemized discrimination, proving that accessibility can overcome systematic imbalances. This drives the primary mission of the FAST team: utilizing cutting-edge technology to enable frontline workers with the requisite institutional knowledge to enable real change.

In the era of major advances in AI-driven tools, we are already noting that benefits are not reaching historically disadvantaged communities (European Disability Forum, International Monetary Fund). The advent of multi-modal generative models has revolutionized the fields of natural language and vision AI. We are seeing influences in education and academia at the college level and beyond in more privileged communities, but the disconnect is stark for disadvantaged communities due to paywalls or unfamiliarity with the new tech. The problem is unique and hard to solve because the impacts are global and pandemic, but a solution has to be local and grassroots-driven. 

We employ cutting-edge multi-modal generative models, in conjunction with open-source and free architecture to make the process of using tools like large language models, cloud-based secure databases, and web interfaces possible for our education partners. We spend significant time understanding the nuances of a classroom environment. Our current prototype for a chatbot-based tutor consists of two web pages: a student portal and a teacher portal. The teacher or administration portal can configure a large language model for a specific education task. This includes adding lesson plans, quiz questions, commands, links to share, as well as safety resources. 

The student portal is completely independent of the teacher portal, which allows a teacher to configure a bot once for an entire class and have all students log onto a simple URL to start their specially created lesson. At the backend, these two portals are connected through a cloud-based database, which also allows for additional file-sharing or student performance tracking. We have found our AI-bot to be able to answer questions, correct misconceptions, drive conversation back to the topic, use the Socratic method, and be affirmative and encouraging. 

Other features we have or plan to implement include image integration (using multi-modal models), text-to-voice, and voice-to-text for accessibility. We are completely open-source, community-powered, and hosted on free resources, making us accessible to groups that are working on a tight budget such as non-profits. Our websites are powered by Streamlit, data storage by Google Firestore, and we use OpenAI's GPT-4 as our AI backend.

The FAST team is on its way to enabling change in the US, India, and Bangladesh by consulting with local groups that serve their respective communities. For instance, demonstrating initial prototypes with local high schools, education outreach programs, and vocational programs in Bangladesh have met great enthusiasm and reviews, with a strong interest in building a solution that is specially engineered for each group's own needs.

github.com/shubhagrawal30/FAST

Student Portal

Teacher Portal (with Demos!)

(you might have to re-wake the prototypes: the hosting website saves crucial environmental resources by turning the app off after some inactivity)

We can answer this question in two parts: we envision that FAST will serve both students and young professionals in historically disadvantaged communities, as well as the teachers, organizations, and volunteers who work with them in educational programs. 

For the former, we envision possibilities ranging from classroom help to at-home access to a robust tutor to co-curricular endeavors such as olympiads or optional coursework or tests. From our own and mentees' experiences, we realize the power of having an on-demand tutor that is extremely knowledgeable and conversational. Additionally, in resource-limited classrooms, the option of assistants in the form of computer-powered bots that also understand the goals of a lesson would be revolutionary. This connects to the latter group too. Aiding volunteers and teachers who work with disadvantaged or excluded groups will help in situations where they are overwhelmed by the limitation of resources. This also aids in retention of competent professionals in the field.

Our team is a multi-faceted, multi-background, diverse group of people. We originate from the communities we are currently working with - growing up, studying, and forming deep connections in the three countries we are currently exploring with our outreach partners. We have noted that our team's diversity has been a great strength for our program, with each of us contributing not only from our professional spaces but also from our lived experiences.

We are also cognizant of our place in our solution: we see ourselves as professionals who happen to have experience with education outreach/policy, generative artificial intelligence, and current technology and coding, but our education partners (school teachers, vocational program workers, outreach volunteers) are the ones with the real-world experience and institutional knowledge to understand the needs and ambitions of their communities. While our current prototype is designed to demonstrate the potential applications and power of our solution, we are now collaborating directly with these frontline workers to design versions of our interface specifically geared for a community's requirements. For instance, this could manifest in the form of a non-English language integration for certain students, or text-to-voice for accessibility reasons.

We have chosen the "prototype" option above, as we have moved beyond the "concept" stage by building a robust version of our tutor, but have not risen to the level of "pilot" as we are still working with our outreach partners to have them use our solution within their community. We have, of course, been using our prototype to demonstrate and pitch our suite to partners, and have tested the tool extensively both internally as well as through our partner executive teams. In concrete numbers, on the order of dozens of people have tested our prototype in anticipation of further use within their communities. 

We have built the teacher and student portal, as well as the corresponding backend (cloud-based storage), as described above. Our websites are powered by Streamlit, data storage by Google Firestore, and we use OpenAI's GPT-4 as our AI backend. The teacher page allows an administrator to configure the tutor bot, while the student page allows several students to then use the tutor. In our experiments, we have found our tutor to be able to use links, commands, lesson plans, the Socratic method, and encouraging affirmations.

github.com/shubhagrawal30/FAST

Student Portal

Teacher Portal (with Demos!)

Some Fun Slides

We are applying to understand our to elevate our platform to a wider audience. As mentioned before and thoroughly apparent from our mission, we are a non-profit initiative that aims to help professionals utilize the best that tech and AI have to offer. The support we need is primarily non-monetary, in the form of understanding the market and cultural barriers that prevent us from being a sustainable solution.

Additionally, our team includes professionals who are volunteering their time for this project, as they are passionate about the possibilities and applications for this solution. We at FAST are great at the parts of the project we have contributed to, but our mission would be immensely served by having mentorship on possible applications beyond what we are envisioning currently (given we realize our potential in education can be vast).

We have four professionals from the education and academic research field who volunteer time for this non-profit initiative.

We have been working part-time on this project for the past half year.