LearnFlow AI — Adaptive Multi-Agent Tutor

Login Page: Secure and user-friendly authentication for personalized learning.
Dashboard: A centralized workspace for courses, AI tutoring, and study materials.
Document Upload: Upload PDFs, slides, notes, and lecture transcripts for AI-powered learning.
Personalized Study Plan: AI identifies learning gaps and recommends the next best learning activity.
Adaptive Practice: Customize topics, difficulty levels, and question counts for targeted practice.
AI Quiz: Interactive questions provide instant feedback, explanations, and source references.
Progress Analytics: Visualize course mastery, topic performance, and knowledge gaps to improve learning outcomes.

💡 Inspiration

Students today learn from a combination of lecture videos, textbooks, PDFs, and presentation slides. However, these resources are often scattered across different platforms, making it difficult to connect concepts, find the right explanation, and understand what to study next.

Generic AI assistants can answer questions quickly, but they may provide information that is not grounded in the student's actual course material. More importantly, they usually do not maintain a detailed understanding of a student's strengths, weaknesses, misconceptions, and learning progress.

We wanted to build a learning companion that understands both the course material and the learner.

🚀 What We Built

LearnFlow AI is a multi-agent AI study companion that transforms lecture videos, textbooks, and slides into a source-grounded knowledge base and uses it to deliver personalized tutoring and adaptive assessments.

The system follows a continuous learning loop:

Learn → Assess → Analyze → Personalize → Learn Again

Students can upload their course materials and interact with them through a source-grounded AI tutor. The system identifies topics, concepts, subtopics, and prerequisites while preserving the original source location such as a textbook page, slide number, or video timestamp.

When a student asks a question, LearnFlow AI retrieves relevant course content and generates an answer grounded in those sources. Instead of simply providing an answer, the system can point the student back to the exact material used to generate the response.

The platform also generates adaptive assessments including MCQs, short-answer questions, and numerical problems. Questions are tagged with their topic, difficulty, and source, and are verified before being presented to the learner.

🧠 Multi-Agent Architecture

LearnFlow AI uses specialized AI agents coordinated through a multi-agent workflow.

1. Multimodal Ingestion Agent

Processes lecture videos, textbooks, slides, images, and diagrams.

2. Knowledge Organization Agent

Identifies topics, subtopics, concepts, and prerequisites and connects them to their original sources.

3. Grounded Tutor Agent

Answers student questions using the course knowledge base and provides source citations.

4. Assessment Generation Agent

Creates adaptive MCQs, short-answer questions, and numerical problems based on the selected topic and learner level.

5. Question Verification Agent

Checks generated questions and answer keys to reduce incorrect or unsupported assessments.

6. Misconception Analysis Agent

Analyzes incorrect answers to identify possible conceptual misunderstandings instead of simply marking an answer as wrong.

7. Learner Model Agent

Maintains topic-level mastery estimates and learning history across assessments and tutoring interactions.

8. Personalization Agent

Uses the learner's current mastery and weaknesses to recommend what the student should learn or practice next.

These agents work together rather than operating as isolated AI tools.

🔄 Adaptive Learning Loop

The core of LearnFlow AI is a continuous feedback loop.

A student learns a concept → takes an assessment → receives feedback → the system analyzes mistakes → learner mastery is updated → the next learning activity is personalized.

For example:

Student struggles with Recursion

↓

The system detects repeated errors

↓

A possible misconception is identified

↓

The tutor provides a targeted explanation

↓

The system generates appropriate practice questions

↓

The student's mastery estimate is updated

↓

Future questions and recommendations adapt to the new learning state.

This allows the system to move beyond static question generation toward personalized learning.

📚 Source-Grounded Learning

A major focus of LearnFlow AI is keeping generated responses connected to the student's actual course material.

Each knowledge unit maintains its source information, such as:

Textbook page
Slide number
Video timestamp
Topic and concept

When possible, tutor responses are accompanied by these references so students can verify the explanation against their own material.

If a question is not covered by the uploaded course material, the system can clearly indicate that the information is outside the available source rather than presenting unsupported information as course content.

🎯 Personalization

Instead of giving every student the same sequence of questions, LearnFlow AI maintains a learner model.

For example:

Topic	Mastery
Sorting	90%
Trees	72%
Graphs	55%
Recursion	40%

If the system observes that a student is struggling with Recursion, it can prioritize that topic and provide targeted explanations and assessments.

The learner model is updated as the student interacts with the tutor and completes assessments.

🛠️ How We Built It

The platform is designed around three major layers:

Multimodal Knowledge Layer

Ingests and processes course materials and preserves source information.

Multi-Agent Intelligence Layer

Uses specialized agents for tutoring, assessment, verification, misconception analysis, learner modeling, and personalization.

Interactive Learning Layer

Provides the student dashboard, source-grounded chat, assessments, progress tracking, and personalized recommendations.

A shared orchestration layer coordinates communication between agents and maintains the overall learning workflow.

🧪 Evaluation

We plan to evaluate LearnFlow AI across both grounding quality and personalization effectiveness.

For the knowledge and retrieval pipeline, we evaluate:

Faithfulness
Answer relevancy
Context precision
Context recall
Citation/source accuracy

For adaptive learning, we evaluate:

Topic mastery improvement
Personalization across sessions
Question repetition rate
Misconception detection
Quality of recommendations

We will use a team-created evaluation set containing questions with known source locations as well as questions that fall outside the uploaded course material.

💻 What We Learned

Building LearnFlow AI showed us that creating an educational AI system is not simply about generating good answers.

The system must also understand where the answer came from, whether the question is supported by the course material, how well the student understands the topic, and what the student should learn next.

The multi-agent architecture helped us separate these responsibilities into focused components while allowing them to work together as one learning system.

🔮 Future Improvements

We plan to extend LearnFlow AI with:

Visual course maps showing topics and prerequisites
Automatically generated revision material
Flashcards and audio study briefs
Personalized study schedules
Mixed-language tutoring
Voice-based tutoring sessions
More advanced learner-modeling techniques

Our long-term goal is to make LearnFlow AI a reliable study companion that adapts not only to what the student asks, but also to what the student understands.