PassClass

Inspiration

As third-year university students we face an endless stream of lectures, provided to us by seemingly vindictive lecturers who appreciate nothing more than filling their lectures with irrelevant content. There needs to be a better way to consume lectures than having to sit through hours of mind-numbing, boring content.

What it does

In order to combat this issue, we at PassClass have developed an application that not only converts lectures and videos to digestible content but also gives the user the ability to test their knowledge with our in-built quizzing function. Furthermore, we have the capacity to upload an entire semester's worth of lecture slides and its information to bolster our LLM’s outputs. Whilst generative AI has made it easier for students to summarise content, the information provided by the AI can be subject to error. It is commonplace for users of Chat-GPT, for example, to become frustrated by its somewhat nonsensical answers. Hence our solution partners with Redactive-AI employing RAG to ensure reliable and accurate information retrieval, making sure that the information it provides is coming from the lecturer rather than Chat-GPT itself. In addition, it ensures that open models such as Chat-GPT do not directly have access to lectures, which is propriety information held by universities. For the future we see this product levelling the education industry, allowing any individual to not only access the lecture but be able to understand and internalise any type of content the user would like. We see our product as the forefront of the education revolution brought upon by generative technologies.

How we built it

After developing the idea, we quickly needed a way to somehow transform video files to text transcripts as well as lecture slides, we settled upon the following technologies: AWS S3, AWS Textract, AWS Transcribe and API Gateway. With this content, we publish it to Atlassian’s Confluence acting as a makeshift database allowing lecture content and even individual notes to completely form the information that the LLM, in this case Open-AI’s ChatGPT, provides to the user. The information, before passing through to ChatGPT is first chunked by Redactive-AI grabbing all the relevant information through semantic search from our database in Confluence. In the front-end we have developed a simple but highly intuitive website made with Flask that enables users to submit their video files as well as lecture slides or even tutorial notes.

Challenges we ran into

Whilst we have had experience with making products before, we still had some issues with dealing with a wide range of API calls. For example, ChatGPT’s API proved somewhat problematic where the API key we used simply did not work for different machines. Additionally, as Redactive AI is a new product, our team had to work with some of the Redactive team to completely understand it and ensure we were using it to the best of its capabilities. Some of our team was relatively new to AWS services and hence, the beginning process we faced some difficulties with Boto3 and linking our process with the S3 buckets.

Accomplishments that we're proud of

As a team we are incredibly new to hackathons and the entire development process in general. Hence, by having a working minimal viable product is a massive achievement, having a working product is something that we are all incredibly happy about. We are also proud of what we have adapted to and learned in these three days. Many of us are new to AWS technologies and web development, and hence given the short time provided to us it is an amazing achievement to learn and deploy it to our local machine. These skills are something that will remain valuable to us long after UniHack has concluded.

What we learned

Throughout this whole process each person of our team was forced to take upon new technologies in order to develop PassClass.

Redactive AI and LLM: Learning Redactive AI required us to deal with a completely new product, and hence instead of the usual process of being able to Google a particular feature we had to directly interact with the Redactive AI team and truly understand their API and its specific technologies. Additionally, by talking with one of the founders of Redactive AI and their engineers we were provided with a direct look into a start-up’s work flow and process. Additionally, we learnt more about how the Chat-GPT API functions and how to connect it with Redactive.

Flask and Web-Development: As a team we have limited web-development knowledge and thus we were forced to tinker with HTML, CSS and JS to not only make a presentable product but also have it interact smoothly with the back-end AWS services and Python.

AWS Services: Whilst having some experience with AWS we nonetheless had to look deeply with Boto3's documentation and figure out how we connect our services to our website. Uploading files to S3 buckets, transcribing video lectures or taking text from lecture slides, whilst seemingly simple was still an exciting new experience especially for those on the team with limited AWS experience. text

What's next for PassClass

We see this application as not only a tool for students but for educators as well, where they can place tutorial notes or even their own summaries for the database to use. We further see a use case of our product to aid those with special needs. For example, we look at a service to translate lecture slides and notes from text-to-speech aiding the visually-impaired. This product is also perfect for those with attention-deficit disorders who are simply unable to sit down with monotonous lecture on hours upon end. In the not-so-distant future we see teachers in primary and secondary schools giving struggling students this tool to ensure that they no longer must fall behind. Whilst teachers are busy with other pupils, students can test themselves and query our program to directly answer the problems the student has.