Our Story
💡 Inspiration
Learning is not the same for everyone. While building and studying software projects, we noticed that students often spend a lot of time searching through PDFs, lecture notes, presentations, and other course materials to find a simple answer. Generic AI tools can answer questions, but they may not always stay within the student's actual course material or explain concepts according to the student's level.
This inspired us to build LearnLens AI, a personalized AI tutoring system that learns from a student's own course material and provides grounded, adaptive learning support.
🚀 What We Built
LearnLens AI turns a student's course content into a personalized knowledge base.
Students can upload learning resources such as PDFs, PowerPoint presentations, Word documents, images, and video content. The system processes these resources, extracts useful information, and creates searchable knowledge chunks.
The AI tutor then uses retrieval-based generation to answer questions using the student's uploaded material. Instead of giving an unsupported answer, LearnLens AI provides source-based citations so students can understand where the answer came from.
The system also supports personalized assessment. It can generate questions from the available course knowledge, validate them, track the learner's performance, and estimate topic-level mastery.
🧠 How We Built It
We built the prototype using Python and FastAPI with a lightweight SQLite database.
The main pipeline includes:
- Course creation and authentication
- Multimodal document ingestion
- Text extraction and processing
- Knowledge-base creation
- Hybrid retrieval using BM25 and hashed dense representations
- Reciprocal-rank fusion and query-coverage reranking
- Grounded AI responses with citations
- Out-of-scope question detection
- Personalized MCQ generation and verification
- Learner mastery tracking using Bayesian Knowledge Tracing
- Recommendations based on weak learning topics
The system is designed to work even without an external LLM API key by using extractive answers and rule-based fallbacks. When an LLM is available, it can provide richer explanations and question generation.
📚 What We Learned
This project helped us understand that building an AI tutor is much more than connecting an LLM to a chat interface.
We learned about:
- Retrieval-Augmented Generation (RAG)
- Multimodal document processing
- Citation validation
- AI response grounding
- Question generation and verification
- Bayesian Knowledge Tracing
- Personalized learning systems
- FastAPI backend development
- Database design
- AI evaluation and reliability
One of the biggest lessons was that an AI system should know when it does not know something. For an educational application, preventing unsupported answers is just as important as generating useful answers.
🛠️ Challenges We Faced
One of our biggest challenges was making the tutor stay grounded in the student's actual course material. A simple keyword search was not enough, so we combined multiple retrieval techniques and added reranking and citation validation.
Another challenge was generating useful assessment questions without producing incorrect or duplicate questions. We therefore designed verification steps to check whether questions are supported by the available source material.
We also had to design the learner model so that it could work with limited student interaction data. We chose Bayesian Knowledge Tracing because it provides an interpretable way to estimate mastery for individual topics.
Building the project as a beginner team also meant dealing with setup, dependency, database, authentication, and deployment challenges along the way.
🌟 Our Vision
LearnLens AI is not intended to replace teachers. Our goal is to build an AI learning companion that helps students understand their own course material more effectively.
We envision a system where a student can simply upload their course resources and receive a personalized learning experience — including explanations, citations, practice questions, mastery tracking, and recommendations based on their individual learning needs.
LearnLens AI — Your course. Your knowledge base. Your personalized AI tutor.
Built With
- adaptive
- artificial-intelligence
- bayesian
- education-tecnology
- fastapi
- generative-ai
- knowledge
- learning-ai
- llm
- machine-learning
- multimodel-ai
- natural-language-processing
- personalized
- python
- rag
- sqlite
- tracing
- tutor
Log in or sign up for Devpost to join the conversation.