Inspiration
This idea originally came from my frustration of failing to find a book that would best teach myself OOP programming. Later I find out that chatbots made learning easy to start, but they rarely remembers what I had actually demonstrated, or let you pick up where you left off without telling it again. It gets a rough idea, yes, but not as good as I wanted. So I decided to create this product.
What it does
Tell it what you want to learn (or bring your own pdf materials), and it creates a structured course for you. It breaks the subject into individual skills, gives you targeted practice, and records what you actually demonstrate rather than simply assuming you understand. After each attempt, it uses your performance, confidence, and reliance on hints to adjust the next question, support level, feedback, and review.
How we built it
Idea: Mine. Learning principles: Codex 'read' the book I provided. Prototype and UI: We built it together, with Codex writing all the code.
Challenges we ran into
Distinguish the LLM from this learning engine. Chatbots are good teachers a lot of times already. So, in order to make something different, consistent learning principles and personalized evidence building will make a difference here.
Another challenge was resisting familiar engagement mechanics. I initially wanted to add rewards, streaks, and stronger motivational incentives. I already figured out the form: a nostalgic farm! However, the learning-science research I reviewed cautioned against relying on external rewards as a lasting driver of learning, especially when they distract from mastery itself. As a result, the MVP uses small skill stars only to reflect independently demonstrated evidence. The learning journey is optional and contains no streaks, guilt, or pressure to return every day.
Accomplishments that we're proud of
The showcasing of coding examples. How do people learn coding without typing and running them? I "fought" with codex and now we have this interface designed. Cool to see what happens when I change the code!
What we learned
A lot of things in the world are counter-intuitive. Like drives for study best come from only internal. Codex is powerful and not judgy. Finally, keeping the LLM behind a modular boundary made the product easier to reason about, test, and adapt to different models. Just learnt that from another book!
What's next for Personalized Learning Engine
Adding more meaningful books. Codex reads much quicker than I do. And based on my structure, it does not forget and it minimizes hallucination. If a user have a specific loved book, allow that book to weight more in the designing stage. Next, I want to strengthen course generation from learning goals and uploaded materials, improve validation of AI-generated exercises, and test the Learning Engine with real learners. Future versions could support more programming languages and subject types, better review scheduling, richer source grounding, multiple learners, and secure cloud synchronization.
Log in or sign up for Devpost to join the conversation.