Learn-it HCL — The Next Generation of Personalized AI Learning

Inspiration

Most traditional learning platforms focus on course completion rather than actual understanding. Watching a video, completing a lesson, or passing a quiz does not necessarily mean a student has mastered a concept or will remember it later.

We were inspired by a fundamental question: What if a learning platform could understand what each student knows, predict what they might forget, and recommend exactly what they should learn next?

This inspired us to develop Learn-it HCL, an evidence-based AI-powered learning platform that transforms traditional education into an adaptive, personalized learning experience.

Our vision is to make learning more efficient, measurable, engaging, and accessible by combining artificial intelligence, machine learning, cognitive science, and educational technology.

What it does

Learn-it HCL is an intelligent learning ecosystem designed to understand learners' abilities, identify knowledge gaps, personalize educational content, and continuously monitor learning progress.

Key Features

• Bayesian Knowledge Tracing (BKT): Estimates the probability that a student has mastered a particular skill based on learning activities and assessment performance.
• Adaptive Assessments: Uses Item Response Theory (IRT) and Computerized Adaptive Testing (CAT) to select questions based on individual ability.
• Intelligent Spaced Repetition: Uses FSRS-inspired memory scheduling to recommend reviews when previously learned concepts may be forgotten.
• AI Course Generator: Uses LangGraph and large language models to generate structured course roadmaps, lessons, and assessment content.
• Personalized Learning Paths: Identifies skill gaps and recommends appropriate learning activities based on goals, prerequisites, and current mastery.
• AI Tutor: Provides conversational learning assistance, explanations, and contextual guidance.
• Neural Video Recommendations: Uses a PyTorch-based neural ranking model to recommend educational videos based on learner characteristics.
• Dropout Risk Prediction: Analyzes learning activity, engagement, and assessment performance to estimate disengagement risk.
• Gamification: Incorporates experience points, achievements, streaks, and learning milestones.
• Teacher and Administrator Dashboards: Provides course management, learner monitoring, analytics interfaces, and role-based administration.

The central principle of Learn-it HCL is simple:

Learn what you need, practice what you struggle with, and review before you forget.

How we built it

We designed Learn-it HCL using a full-stack architecture that separates the user-facing learning experience from specialized AI and machine-learning services.

Frontend Development

We built the web application using Next.js, React, TypeScript, Tailwind CSS, and shadcn/ui.

The frontend includes personalized dashboards, course exploration, adaptive quizzes, skill mastery visualization, spaced-repetition review queues, learning pathways, gamification features, and administrative interfaces.

Backend and AI Services

We developed a Python FastAPI backend to handle machine-learning inference, adaptive assessments, personalized recommendations, AI tutoring, and course generation.

We implemented several specialized learning algorithms:

1. Bayesian Knowledge Tracing: Probabilistic estimation of individual skill mastery.
2. FSRS Spaced Repetition: Memory stability and retrievability estimation for personalized review scheduling.
3. 2-Parameter Logistic IRT: Ability estimation and question difficulty modeling.
4. Computerized Adaptive Testing: Dynamic question selection using Fisher information.
5. Logistic Risk Prediction: Behavioral risk estimation for student engagement monitoring.
6. Neural Video Ranking: A lightweight PyTorch multilayer perceptron for educational content ranking.
7. Recommendation Pipeline: An 11-stage system combining learner profiles, goals, skill gaps, prerequisites, ranking, and explainability.

Generative AI

We incorporated LangGraph to orchestrate AI-driven course generation through multiple stages, including roadmap creation, educational material generation, and assessment generation.

The architecture supports LLM integrations through Groq and OpenAI-compatible APIs.

Database and Infrastructure

We used PostgreSQL for structured educational data, Prisma ORM for application data access, and SQLAlchemy for Python-based database operations.

The deployment architecture uses Vercel for the frontend and Microsoft Azure Container Apps for backend services, with Docker and GitHub Actions supporting containerization and deployment workflows.

Challenges we ran into

1. Measuring real knowledge

One major challenge was distinguishing actual knowledge mastery from basic quiz accuracy.

A correct answer does not always indicate understanding, and an incorrect answer does not always indicate a lack of knowledge.

We addressed this modeling challenge by implementing Bayesian Knowledge Tracing, which accounts for guessing, mistakes, and learning progression.

2. Building adaptive assessments

Traditional quizzes deliver the same questions to every student.

Developing an adaptive testing system required combining ability estimation, question difficulty, discrimination parameters, and dynamic question selection.

We approached this using Item Response Theory and a Computerized Adaptive Testing engine.

3. Predicting memory retention

Learning effectiveness depends not only on understanding information today but also on remembering it later.

Designing an intelligent review system required modeling memory stability, difficulty, and retrievability.

We implemented an FSRS-based scheduler to estimate review timing.

4. Combining multiple AI models

Integrating probabilistic knowledge tracing, adaptive testing, recommendation algorithms, neural ranking, and generative AI into one application presented architectural challenges.

We organized these capabilities into separate modules with APIs and shared learner-state information.

5. Explainability and personalization

Recommending a lesson is useful, but students should also understand why it was recommended.

We designed the recommendation engine to produce explanations alongside personalized learning suggestions.

6. Multi-user architecture

Supporting students, teachers, administrators, and organizations required role-based access control, structured

Built With

• azure
• azure-container-apps
• docker
• fastapi
• github-actions
• groq
• jwt
• langgraph
• next.js
• node.js
• openai
• pnpm
• postgresql
• prisma
• pydantic
• python
• pytorch
• react
• shadcn/ui
• sqlalchemy
• tailwind-css
• turborepo
• typescript
• vercel