Personalized Tutoring & Adaptive Learning

• 

  Overall Flow

Inspiration

Every student learns differently, but traditional learning systems often provide the same content and assessments to everyone. Students may struggle to identify their weak topics, find reliable explanations, or decide what to study next.

We built Personalized AI Learning & Student Support Platformto make learning more personalized, measurable, and accessible. Our goal is to help students understand concepts, identify learning gaps, and receive guidance based on their learning progress, while also connecting them with relevant government scheme eligibility support.

What it does

The platform combines AI-powered learning tools with student support features in one application.

• AI Tutor: Answers questions using uploaded learning materials and provides source-grounded responses.

• Knowledge Base: Allows students to upload PDF learning materials and retrieve relevant information.

• Adaptive Assessments: Generates topic-based quizzes at different difficulty levels based on the learner's recorded mastery.

• Learner Model: Tracks assessment attempts, accuracy, and topic-level mastery.

• Learning Gap Detection: Identifies topics that need improvement and categorizes active gaps by severity.

• Personalized Recommendations: Prioritizes topics for further practice using learning gaps and mastery information.

• Student Support: Integrates an existing government scheme eligibility predictor to help students explore potentially applicable schemes.

How we built it

We developed the platform using a modular architecture with separate learning and student-support components.

The backend uses Python and FastAPI, with PostgreSQL and pgvector for storing learning documents, text chunks, embeddings, assessment records, and learner progress. PDF content is processed and converted into searchable chunks. Embeddings and vector similarity search help retrieve relevant material for the AI Tutor.

The RAG (Retrieval-Augmented Generation) workflow retrieves relevant learning content before generating an answer. The platform attaches source metadata to responses and avoids generating an AI answer when relevant learning context is unavailable.

The assessment engine generates questions from retrieved learning materials, evaluates submitted answers, updates topic-level mastery, and identifies learning gaps. The adaptive assessment workflow uses recorded mastery to select appropriate question difficulty.

The frontend is built with Next.js, React, TypeScript, and Tailwind CSS, providing dashboards for tutoring, assessments, progress tracking, learning gaps, knowledge management, and student support.

The government scheme eligibility predictor remains a separate support component so that educational recommendations and scheme eligibility decisions are not mixed together.

Challenges we faced

One of our main challenges was ensuring that AI responses remain grounded in the student's uploaded materials rather than relying on unsupported information. We addressed this by integrating vector search, relevance thresholds, source metadata, and explicit no-context handling.

Another challenge was connecting assessment results to meaningful learning progress. We implemented a transparent mastery calculation, active learning-gap tracking, and difficulty selection based on recorded mastery.

Integrating these features into one application also required careful separation between the learning engine and the government scheme support engine. We tested API behavior, validation, error handling, empty states, and responsive UI behavior.

A current limitation is that live AI tutoring and assessment generation require a configured LLM API key and available API credits. Course management also remains limited because a dedicated course-management API has not yet been implemented.

What we learned

Through this project, we gained practical experience with RAG pipelines, vector databases, embeddings, API development, assessment workflows, learner-state tracking, adaptive learning logic, and frontend-backend integration.

We also learned that personalization is more than generating AI responses. A useful learning platform must connect content, assessments, progress, learning gaps, and recommended next steps into a continuous feedback loop.

What's next

Our next steps include completing live LLM-powered testing, implementing dedicated course management, improving learning analytics, and expanding support for additional educational content formats. We also aim to strengthen personalization as more assessment history becomes available.

Our vision:Help every student understand what they know, discover what they need to improve, and know what to learn next.

Built With

• adaptive-learning
• artificial-intelligence
• education-technology
• fastapi
• generative-ai
• large-language-models
• machine-learning
• next.js
• personalized-learning
• pgvector
• postgresql
• python
• rag
• react
• rest
• tailwind-css
• typescript
• vector-search