AITutor

Inspiration

We wanted to build something smarter, an AI tutor that behaves more like a human teacher: observing patterns, remembering mistakes, adjusting difficulty, and continuously improving its teaching strategy based on the student’s performance. What if an AI tutor could actually learn how you learn?

What it does

Understands student intent (practice, explanation, feedback, etc.) Tracks performance history and skill levels over time Adapts difficulty dynamically Provides tailored feedback and encouragement Unlike a normal chatbot, our agent maintains a persistent student model that evolves after every interaction. The more you practice, the more personalized your experience becomes.

How we built it

We designed AITutor using a multi-agent architecture, where each agent has a specialized role: Perception Agent → Interprets student requests and context Memory Agent → Manages student profiles, skill levels, and interaction history Planning Agent → Chooses optimal next actions based on learning state Evaluation Agent → Grades answers and generates structured feedback

The system runs through an adaptive loop: Analyze student input Retrieve learning history Decide next best action Evaluate performance Update skill levels Store memory for future personalization

We built the frontend in Streamlit, used structured JSON knowledge bases for content, and integrated a Ollama to power reasoning and feedback.

Challenges we ran into

Making JSON outputs reliable from LLM responses Handling exercise repetition and properly tracking completed exercises Ensuring student profiles updated correctly across sessions

Accomplishments that we're proud of

Built a fully working multi-agent adaptive system, not just a chatbot Implemented real skill-level progression using performance signals Created persistent student profiles that evolve over time Designed an architecture that can scale to other exam domains beyond IELTS

What we learned

Multi-agent design leads to cleaner, more scalable AI systems Memory and state are critical for meaningful personalization Small details like exercise tracking significantly impact user experience

What's next for AITutor

• Making it scalable for other domains
• To integrate a Meta-Learning Agent so that the it can optimize its own learning algorithm and how to adapt in new environment

Built With

• ollama
• python