Inspiration
The idea for CogniLearn came from my own interest in cognitive psychology. I've spent time reading papers, taking courses, and recently, we were assigned a paper in class on how retrieval practice and feedback outperform passive review for memory retention. It’s well-known in psychology that re-reading notes doesn’t help much — and can even give us false confidence — yet most tools still rely on that.
This got me thinking: what if we built a learning assistant that actually followed those principles?
As someone also interested in computer science, I saw an opportunity to merge these two fields. I brought the idea to the team during the hackathon, and after explaining the science behind it, everyone was excited to bring it to life.
What it does
CogniLearn is a learning assistant powered by AI — but instead of just generating quizzes or notes, it actually guides users through a science-backed study process.
- You input your notes or upload content
- The app summarizes key points using AI
- Then it prompts a free recall attempt before showing anything
- That response is evaluated based on the summary to give meaningful feedback
- After that, it generates a quiz with short answers, MCQs, and explanation questions
- Each answer is judged with cognitive feedback, and missed questions are tracked for later
The focus is on retrieval, reflection, and reinforcement — not passive consumption.
How we built it
We used Python and Flask to build the backend APIs and React for the frontend. The AI model used for content generation and evaluation is Google's Gemini API.
Our team of four split responsibilities — I acted more as the PM and learning architect, focusing on the research-driven flow, prompt design, and interaction logic. Three of us were new to building full-stack apps, so there was a lot of learning on the fly.
The core loop — from content to recall to quiz — was designed to reflect known principles in cognitive science like free recall, interleaving, and elaborative feedback.
Challenges we ran into
- Prompt tuning was difficult. Getting reliable, structured data back from the model took a lot of trial and error.
- Making the responses machine-parseable while keeping them human-readable wasn’t easy.
- Designing the learning flow itself was a challenge — we wanted it to feel guided and meaningful without restricting users too much.
- Since several of us were first-time developers, integrating frontend and backend in a clean way took time and debugging.
- We also struggled with framing the feedback — making it helpful without making users feel punished or micromanaged.
Accomplishments that we're proud of
- Turning solid cognitive science theory into a working, interactive tool.
- Building a full end-to-end stack with backend APIs, AI integration, and a usable frontend.
- Creating a flow that isn’t just convenient — it’s meaningful and grounded in how memory works.
- Staying focused on the goal: helping users actually learn, not just feel productive.
What we learned
- That bringing perspectives outside of computer science — like psychology — can spark fresh, meaningful ideas. Having a broader horizon helped shape a more impactful project.
- We got a solid understanding of how full-stack apps actually work — from Flask endpoints to sending and receiving requests — which we hadn’t fully explored in class.
- We learned how to think like developers, not just coders — by designing with the end user in mind, accounting for edge cases, and making intentional product decisions instead of just building features.
- We saw firsthand how prompt design and data parsing affect product reliability — especially when using AI models.
What's next for CogniLearn - A Science based Learning Approach
We’re excited to keep refining CogniLearn. Here's what we plan to add:
- Cram Mode: A time-sensitive mode that prioritizes recall and fast feedback over spaced learning — useful when students only have a day or two to study.
- Recall Score Tracking: A progress indicator showing what percentage of the content a student successfully recalled — helping them gauge their retention, not just completion.
- Adaptive recall schedules using spaced repetition and prior performance
- Support for interleaving across multiple topics or tests (e.g., midterms for different subjects)
- A better student dashboard to visualize weak points, feedback, and progress over time
- Option for educators to upload content and build guided learning modules
We're committed to building a tool that doesn’t just support studying — but actually helps people learn better using science.
Log in or sign up for Devpost to join the conversation.