My debut full-stack project for neonlearner.com with LLM integration using OpenAI technologies: A dynamic open-ended questionnaire designed to delve deep into users' study habits and gauge the efficacy of their learning strategies.
The project was created as a part of my initiative to make online learning more accessible to students by educating them on the effective learning strategies and independent learning capabilities. This test is a free introductory test for anyone who wishes to see how effective their current learning strategies are. The effectiveness score is calculated by comparing user response with the ideal learning methodologies grounded in the science of learning research. Users can also opt to receive detailed insights into their learning style, coupled with guidance on potential improvements.
Technically speaking, the frontend leverages HTML, CSS, and JavaScript, while the backend primarily utilizes the node.js runtime, hosted on AWS Lambda. The backend also interacts with AWS dynamoDB and SES services to fetch, record and send the required information. The connection between frontend and backend is established by HTTP API, provided by AWS API gateway service. The backend also makes API calls to OpenAI services. The latest OpenAI model at that time, which is GPT-3.5-turbo was used to evaluate user answers and assign a score. The model played a crucial role in this project by allowing to accept open-ended answers from the users and thus, improved the UX. The whole infrastructure is monitored by AWS Cloud Watch that informs me about the errors my clients were experiencing.
I opted for a single Lambda instance deployment to streamline API gateway routing. Rather than setting up multiple routes, the Lambda function processes various request types embedded in the JSON string, decoded within the lambda event handler. While a more modular architecture was feasible, the web application's non-intensive nature negated the need for such complexity. Furthermore, Lambda instance contains various security checks that can not be displayed in any public repository