Ami is an adaptive tutoring system that personalizes learning goals, learning paths, session content, quizzes, and tutoring support around each learner. This repository started from GenMentor (WWW 2025, Industry Track), but the current codebase extends that work into a fuller tutoring platform with persistent runtime state, verified-content grounding, adaptive learner modeling over time, and production-style backend/frontend integration. The system is grounded in two pedagogical frameworks:
Felder-Silverman Learning Style Model (FSLSM): characterizes each learner across four dimensions (active/reflective, sensing/intuitive, visual/verbal, sequential/global) to shape content format and presentation
SOLO Taxonomy: classifies cognitive complexity across five levels (pre-structural → extended abstract) to calibrate content difficulty and quiz depth
Xinping Wang
Digital Transformation and Innovation - UX concentration
UX Engineer
Nellie Le
Digital Transformation and Innovation - UX concentration
Learning Researcher
Tian Lai
Digital Transformation and Innovation - UX concentration
UX Design Lead
Thuy Tran
Digital Transformation and Innovation - UX concentration
Project Manager/Project Coordinator
Tianci Li
Digital Transformation and Innovation - Applied Data Science
Technical and Ethical Framework
Tian Lai
Digital Transformation and Innovation - Applied Data Science
Technical Lead(Multi-agent AI & LLM integration)
Dr. Ali Abbas
CEO Smart Digital Medicine
Adjunct Professor, University of Ottawa
Role in Project: Dr. Abbas provides strategic industry oversight and expertise in intelligent informatics to ensure project relevance. He also defined the project's goal and provided key feedback throughout the project.
Dr. Ismaeel Al-Ridhawi
Associate Professor, Kuwait College of Science and Technology
Technical Advisor, University of Ottawa
Role in Project: Dr. Al-Ridhawi offers specialized guidance on multi-agent AI orchestration and system architecture.
The screenshots below show the current React interface and key adaptive behaviors.
1. Register
New learners create an account with a username and password to get started. After signing up, they are taken directly to the login screen.
2. Login
Returning learners sign in to pick up exactly where they left off — whether that is a learning session, a quiz, or a goal they are working toward.
3. Onboarding
During onboarding, learners choose a learning persona that reflects how they best absorb information, describe their learning goal, and optionally upload a résumé to help Ami understand their background.
4. Skill Gap Indentification
Ami refines the learner's goal into a clear target and surfaces the specific skills needed to get there, grounded in verified course materials and the learner's stated background.
5. Learning Path Personalization
Ami builds a personalized sequence of sessions tailored to each learner's cognitive style and current knowledge level. Sessions are unlocked progressively as mastery is demonstrated.
6. Learning Session and Content Delivery
Each session delivers lesson content, visuals, optional audio, and embedded quizzes in a format matched to the learner's preferred style — whether that means diagrams and worked examples or narrative explanations and structured outlines.
Before presenting a learning path, Ami runs a self-evaluation loop to ensure the plan is coherent, well-sequenced, and appropriate for the learner's level.
7.1. Quizzes
For beginners, quiz questions are calibrated for learners building foundational understanding. For higher levels, quiz questions are designed for those ready to apply and connect concepts across topics.
7.2. Response Assessment
Open-ended responses are evaluated by an AI grader aligned with the SOLO Taxonomy — a research-backed framework that measures how deeply a learner understands a topic.
8. Ami Chatbot Tutor
Ami is available throughout the learning experience as a conversational tutor. Learners can ask follow-up questions, request clarifications, or explore related topics — and Ami draws on the current session content, verified course materials, and the web to respond.
9. Dashboard
The home dashboard gives learners a clear view of their active goal, current progress, and the next recommended step in their learning journey.
10. Learner Profile
The learner profile tracks cognitive progress, learning style preferences, and behavioral signals accumulated across sessions — giving learners and instructors a transparent view of how the personalization is working.
11. Goal Management
Learners can create, switch between, and manage multiple learning goals — making it easy to pursue different topics or return to something set aside earlier.
12. Learning Analytics
The analytics dashboard surfaces progress, skill mastery, session time, and quiz performance — helping learners understand how they are advancing and where to focus next.
13. AI Transparency and Bias Audits
Disclaimers are embedded across the system whenever it uses AI-generated content are audited for inherent biases
14. Verified Course Content
Lessons can be grounded in actual course content, enhancing existing content. References are cited with page number so learners could check the actual source
This project has been a powerful exercise for us in bridging the gap between technical complexity and user-centered design. Together, we successfully integrated multi-agent orchestration with pedagogical theory, proving that high-quality educational AI is about more than just generation; it’s built on grounding, transparency, and trust. Our decision to prioritize verified text and audio over resource-heavy video was a key win, allowing us to deliver a polished and accessible MVP.
While our individual execution was high, the transition to the integration phase taught us that collaboration is where the real work happens. We realized that as the project grew, cluttered group chats weren't enough; we needed the structure of asynchronous stand-ups and deliberate dependency tracking in Jira to keep our work visible. These shifts helped us clarify our roles and keep the technical engine aligned with our design goals. Ultimately, our success was built on our willingness to adapt and listen to each other's and stakeholders' feedback. Looking ahead, we’re excited to implement earlier integration checkpoints and more formal handoffs to make our interdisciplinary teamwork even more seamless.