From Answer Keys to Socratic Mentors
Fostering Learner Agency and Self-Regulated Learning through Adaptive AI Scaffolding in Programming Education
AlgoGPT is an AI-powered learning platform for programming education, conceptualised and led by Dr Owen Noel Newton Fernando and supported by an NTU EdeX grant. It supports learning through guided hints rather than direct answers. Using Socratic questioning and a Driver–Navigator learning model grounded in Cognitive Load Theory and Self-Regulated Learning, students alternate between implementing solutions independently, explaining their reasoning, and critically evaluating AI-generated code to strengthen algorithmic thinking, reflection, and problem-solving skills.
SC1007 — Data Structures and Algorithms is a foundational, compulsory course for every undergraduate at the College of Computing and Data Science (CCDS), Nanyang Technological University — a cohort of around 1,000 students each year. It was redesigned under the College's Learn with AI framework for an AI-shaped future, where understanding why code works matters more than simply generating it.
SC1007 Data Structures and Algorithms, incorporating AlgoGPT, was selected as one of ten winners of the 2026 GMA Award for Outstanding Courses — achieving the highest overall evaluation score among the ten award-winning courses, according to communication from the GMA organisers.
SC1007 Data Structures and Algorithms, incorporating AlgoGPT, was selected as one of ten winners of the 2026 GMA Award for Outstanding Courses.
This recognition highlights its innovative approach to personalised learning, independent problem-solving and critical evaluation of AI-generated code.
Evaluated by an international committee of education experts and learners drawn from UNESCO and other leading institutions.
Highest overall evaluation score among the ten winnersAccording to communication from the GMA organisers.
Official Award Announcement →Dr Owen Noel Newton Fernando was selected for NTU’s 2026 John Cheung Emergent Technology Award, recognising his contributions to technology-enhanced education, including the innovative application of AlgoGPT in Data Structures and Algorithms.
Navigator Mode's flawed-AI-partner design was published at ACM DIS 2026, and AlgoGPT was featured by NTU InsPIRE in October 2026.
NTU's College of Computing & Data Science is embedding AI into how computing is taught, practised and assessed through its Learn with AI framework. In SC1007, students explore core data structures and algorithms in an AI-supported environment that offers guided hints and feedback — never finished solutions — as they solve problems. Practice is dynamically generated to each student's skill level, so a cohort of over a thousand learns at its own pace while building the judgement to evaluate AI-generated code.
“Artificial intelligence is a multiplier. But if the multiplicand is zero, the outcome is zero.”
Every problem moves a student through three modes of thinking — the structured approach at the heart of SC1007's redesign. AlgoGPT supports each stage with Socratic questions, never finished code.
Students put their algorithmic logic into words before any code is written, externalising the mental model so gaps surface early.
Students write and refine the solution themselves, receiving tiered Socratic nudges only on request — direction, never the answer.
Students review AI-assisted output critically, verifying it against the fundamentals — the skill that matters most in an AI-shaped world.
Each two-hour lab is built around the Zone of Proximal Development — students first build reasoning unaided, then extend it with adaptive AI support. The same AI-ON / AI-OFF principle carries into assessment: independent mastery is always verified separately from effective tool use.
Students design and analyse linked lists, stacks, queues, trees and graphs by hand — developing problem decomposition and debugging with no GenAI in the room.
Students pair with AlgoGPT in Driver and Navigator modes while Teaching Assistants act as meta-coaches — guiding responsible, reflective use of every piece of AI feedback.
AlgoGPT supports NTU2030’s emphasis on personalised, AI-supported education. The course follows a deliberate sequence: AI-OFF conceptual learning builds foundations, AI-supported practice provides guided application, and independent assessment requires students to show their own competence. Each of NTU2030’s four dimensions maps to a specific part of the platform.
Understanding what AI is, how it works, its capabilities and societal impact.
Deliberately flawed AI code makes the strengths and failure modes of generative AI visible in every lab.
Engaging with AI tools effectively, ethically and responsibly.
In Navigator Mode, students review AI-generated code, decide whether to accept or reject it, and identify, explain and correct the flaw.
Collaborating with AI to enhance teaching, creativity, and learning experiences.
In Driver Mode, students write their own code while a four-tier Socratic ladder guides them without ever giving runnable solutions.
Developing meta-intelligences underpinning human wisdom, ethics and emotional understanding that AI cannot replace.
Reflection and No-AI practice build independent judgement, self-regulation and problem solving.
Standard large language models act as a black box: they hand over working solutions instantly, letting students bypass the very struggle that builds deep understanding. Learning theory is clear that durable knowledge requires effortful, active construction — and frictionless assistance short-circuits it.
Immediate, complete solutions. Speed without comprehension — and a learner who never owns the logic.
Structured friction by design. Hints, questions and deliberately flawed code that force reasoning — turning answers into earned insight.
AI systems can generate code remarkably quickly, but understanding why that code works remains fundamental. In our courses, students learn to use these tools responsibly while still making sound design decisions, reasoning through problems, and verifying the correctness of their solutions.
Alongside its four learning modes, AlgoGPT provides personalised and adaptive learning support through several integrated features. Students can practise at their own pace, assess their understanding and receive personalised support based on their progress.
Over 400 programming questions covering fundamental and advanced topics in Data Structures and Algorithms.
Over 300 MCQs to assess understanding, reinforce key concepts and identify knowledge gaps.
Personalised practice recommendations based on MCQ responses, laboratory submissions and programming attempts.
Individual profiles that help students monitor their progress, understand their proficiency and identify areas requiring improvement. Each profile begins with a ten-question adaptive assessment and is updated as students learn.
Dashboards that provide students and educators with insights into learning progress, common difficulties and areas requiring additional support.
A closed lab system with access codes, enrolment, and session tracking — built around the professor → TA → student hierarchy.
These features support students in planning their learning, monitoring their understanding, adjusting their strategies and reflecting on their progress. Educators can also use learning analytics to identify common misconceptions and provide more targeted support.
AlgoGPT supports four complementary learning modes that balance independent problem-solving with structured AI assistance.
Students solve programming problems independently without AI assistance, applying their knowledge and developing confidence in their own problem-solving abilities. This mode provides opportunities to practise and assess independent understanding.
Students receive personalised hints, explanations and guiding questions to help them understand programming concepts and overcome difficulties without receiving complete code solutions. The AI provides assistance while encouraging students to develop their own answers.
Students write their own code while AI provides Socratic questions and structured guidance. This mode supports productive struggle by encouraging students to reason through problems, make decisions and refine their solutions rather than relying on AI to write code for them.
Students explain their intended approach and evaluate AI-generated code containing deliberately introduced errors. They identify mistakes, assess the proposed solution and decide whether to accept or reject it. This mode encourages students to question AI outputs, investigate errors and develop independent reasoning and code evaluation skills.
Together, these modes provide different levels of AI assistance while keeping students actively involved in the learning process. Driver and Navigator Modes particularly emphasise learner agency, productive struggle and critical engagement with AI-generated solutions — powered by two specialised, role-reversing agents explored below.
Students write their own code; the AI intervenes only on a student-initiated request. Rather than offering fixes, the Driver decomposes feedback into four escalating tiers — functioning like a senior instructor who scaffolds through questions, never supplying runnable code.
Students receive personalised hints, explanations and guiding questions to help them understand programming concepts and overcome difficulties — without receiving complete code solutions. For research, it also serves as a within-system baseline that isolates the effect of Driver and Navigator Modes.
Students solve programming problems independently without AI assistance, applying their knowledge and developing confidence in their own problem-solving abilities. Automated test cases still give performance feedback independently of generative AI.
AlgoGPT is built on a stateless backend with LangGraph agent orchestration and Azure-hosted GPT-4o models providing pedagogically constrained reasoning. Each agent has one job; together they manage cognitive load end to end.
Runs the diagnostic quiz, classifies mastery, and generates each student's personalised learner profile.
Inspects submissions to detect misconceptions and the specific reasoning gap behind a failed attempt.
Guards the Socratic protocol — ensuring feedback never leaks runnable solutions to the student.
Delivers real-time, non-directive feedback through reflective prompts and scaffolding strategies.
Sequences problems and difficulty, moving learners along a personalised topic progression.
Grounds every agent response in the exact problem statement and course knowledge base.
Across a 14-week semester, AlgoGPT's friction modes produced significantly higher-quality behavioural patterns than standard AI assistance.
Lower "frictionless" passing in friction modes reflects deeper cognitive effort — not failure.
26% ran to time expiry — an "immersion effect", not abandonment.
Driver mode raises iterative engagement; Navigator mode shows the "efficiency of insight" — fewer attempts, deeper pre-submission analysis. Structured modes roughly double time on task, which suggests sustained engagement (time alone does not prove deeper understanding).
| Metric | Normal · Baseline | Driver · Socratic | Navigator · Evaluator |
|---|---|---|---|
| Eventual success rate | 88.0% | 58.6% | 55.0% |
| Mean attempts per session | 3.07 | 3.24 | 2.36 |
| Attempts to first accept | 2.46 | 2.66 | 1.80 |
| Mean tests passed / submission | 3.51 / 10 | 5.13 / 10 | 3.57 / 10 |
| Mean session duration | 10.9 min | 21.6 min | 20.1 min |
Across the 14-week semester, 35,433 submissions were processed without meaningful attrition. Lab-level success climbed steadily as students adapted to friction-first practice — engagement increased rather than eroded.
“Driver and Navigator modes are quite engaging—it really forces me to think instead of just copying answers.”
“Navigator Mode helped me understand the logic better because I had to explain it clearly to the AI.”
“The hints guide me without giving the answer, which improves my confidence.”
“Explaining my reasoning for each MCQ really helped me understand the concepts better.”
AlgoGPT draws on three complementary educational principles to support meaningful learning with AI.
Students take an active role in managing their learning through four processes: planning, monitoring, control and reflection. They identify areas for improvement, monitor their understanding, adjust their problem-solving strategies and reflect on feedback — supported by personalised learner profiles, adaptive practice recommendations and learning analytics.
Programming requires students to understand problems, select appropriate algorithms and translate their reasoning into code. Structured hints, guiding questions and adaptive scaffolding help manage these demands — offering support when needed rather than complete solutions, to reduce unnecessary load while preserving the mental effort required for learning.
Students work through meaningful challenges instead of immediately relying on AI-generated answers: investigating difficulties, testing their reasoning and revising their solutions. In Navigator Mode, deliberately introduced errors create opportunities to identify and explain mistakes. The aim is to make struggle productive through appropriate guidance and feedback, rather than leaving students frustrated or dependent on AI.
Overall mean 3.39 / 5.0; all 12 items significantly above the neutral midpoint. "Checking understanding" gave the study's largest effect (t = 6.97).
AlgoGPT profiles every learner and detects struggle at a fine grain, so 24/7 Socratic coaching reaches a diverse cohort once shaped by access to private tutoring.
Week-2 profiling quiz — N = 760 students classified by the Learner Profiling Agent.
A mostly-beginner cohort — the profile that benefits most from progressive, confidence-building scaffolding.
A 9-level ordinal scale moves beyond code correctness to capture cognitive difficulty.
A lightweight, web-based platform — no high-end hardware, no installs. World-class scaffolding on any device.
Immediate, personalised feedback for learners who cannot afford private academic support — available at every hour.
Scaffolded support for struggling learners; critical evaluation for advanced ones — diverse entry points, one system.
A stateless architecture scales from ten students to ten thousand on standard infrastructure. The pedagogy is domain-agnostic, and the roadmap turns today's uniform friction into tomorrow's adaptive scaffolding.
Stateless sessions with no server-side memory overhead scale effortlessly on Azure container infrastructure and a React frontend.
The tiered-nudge protocol is tied to structured reasoning, not a language — ready for Systems Programming, Maths, Physics and beyond.
Automated, AI-blind test cases keep grading objective; submission logs let instructors detect unproductive frustration early.
The Driver–Navigator framework fits any course with practical tasks, identifiable errors and clear evaluation criteria. Adapting it needs course-specific diagnostic questions, task banks, guidance prompts, representative flaws and automated checks — packaged as reusable templates, rubrics and instructor orientation.
Calibrating the intensity of "friction" to a student's real-time struggle, so productive struggle never tips into unproductive confusion.
Mermaid.js integration to render data-structure and algorithm visuals inside the feedback loop, anchoring abstract concepts.
Whisper-based models to let students give and receive feedback by voice — humanising the pair-programming experience.
Pre/post assessments and difficulty-controlled comparisons to move from evidence of engagement to measured learning gains.
Drawing on more than ten years of teaching Data Structures and Algorithms, Dr Fernando conceptualised AlgoGPT and led its educational design, development and integration into SC1007, supervising PhD, FYP and URECA students who turn pedagogical requirements into platform features. Faculty collaborators Dr Siyuan Liu and Mr Ong Chin Ann work alongside him to deploy and refine AlgoGPT in SC1007. PhD student Mr Yeo Shi Jing investigates how AI can improve, rather than degrade, students’ learning. Current and former FYP and URECA students contribute to its research, development, testing and evaluation, bringing student perspectives into the design of AI-supported learning.
Improving Learning with AI — co-supervised with Prof Jagath C Rajapakse.
We acknowledge the contributions of former students who supported the development and advancement of AlgoGPT.
AlgoGPT also serves as a research platform for investigating how students learn through interactions with generative AI, particularly when AI-generated solutions contain mistakes that encourage critical evaluation, reflection and independent problem-solving.
The peer-reviewed deployment analysed 907 students and 14,958 submissions across 27 DSA problems; the figures throughout this site reflect the expanded post-publication dataset of 35,433 submissions.
Explores how deliberately imperfect AI partners can encourage students to question AI-generated solutions, identify errors and engage more actively in programming problem-solving — the approach reflected in AlgoGPT’s Navigator Mode.
Published by NTU’s Institute for Pedagogical Innovation, Research and Excellence (InsPIRE), the article highlights how AlgoGPT encourages productive struggle, learner agency and independent problem-solving through structured AI guidance rather than complete solutions — and how educational principles can inform the design of AI learning experiences.
We envision a future where AI supports students in becoming more capable, confident and independent learners — combining personalised AI guidance with meaningful opportunities for productive struggle, reflection and critical thinking.
Rather than measuring success solely by what students can accomplish with AI, we focus on what they can understand, explain and solve independently after learning with it.