Undergraduate research

Hints,not answers.

A coding workspace with an AI tutor that guides you, not solves it for you.

What is PACE

PACE is a research prototype. It puts a hint-only AI tutor inside a browser code editor and uses hearts to limit how often students ask. We are studying whether students and instructors accept it.

Pair-programming

Practice real code in Python, C, JavaScript, or C# with a tutor beside you.

A tutor that explains and hints, limited by hearts.

Coding

Write, run, and test your code right in the browser.

Environment

Assignments, tests, and submissions live in one place.

The problem

Learning to code has a tool problem and an AI problem.

What we saw in our own classes, and what the research says, point to three gaps.

IDE

Docs

PDF

LMS

4 apps · 1 assignment

Too many tools for one assignment

Students write code in an IDE, paste it into a document, export a PDF, and upload it to the LMS. None of these tools were built to work together.

Birillo et al., 2024

Write my sum program

Here is the complete solution:

a = int(input())

b = int(input())

print(a + b)

Copied · Submitted · 100%

AI that hands out answers

General chatbots give complete solutions by default. Students can submit code that runs without understanding why it works.

Sun et al., 2024 · Zhai et al., 2024

next step?

why error?

fix it pls

what now?

just answer

again?

∞

Help with no limits

When help is instant and unlimited, students tend to ask before they try. Knowing when to ask for help is a skill that takes practice.

Aleven et al., 2003 · Zimmerman, 2002

Our approach

One place to code, and a tutor that knows when to hold back.

PACE answers each problem with one design decision.

Answers problem 01

One workspace

Read the assignment, write code, run tests in a sandbox, and submit, all in the browser. Scores come from the share of test cases passed.

PythonCJavaScriptC#

Answers problem 02

A tutor that guides, not solves

It reads your code and latest test results, then does four things only: explain code, hint at errors, clarify concepts, and guide problem-solving. It declines requests for full solutions.

Answers problem 03

Hearts make each question count

Each assignment gives you 5 hearts. A question costs one, and one comes back every 30 minutes. Instructors can change both.

+1 in 30 min

How it works

Help that stops short of the answer.

The tutor lives next to the editor, so it can see your code and your latest test results. Here is what it does, and what it won’t.

Four kinds of help, nothing more

Ask the tutor to explain your code, point to where an error likely is, clarify a concept, or suggest your next step. Ask it for the full answer and it declines, then offers a hint instead.

Design based on CS50 Duck (Liu et al., 2024)

Explain my code

Where’s the bug?

Explain a concept

What’s my next step?

Just give me the fixed code.

PACE Tutor

I can’t write it for you, but I can point you there: line 3 adds two values that came from input(). What type does input() return?

Declined: full solution

Hints get more specific as you spend hearts

Your first question gets a conceptual nudge. Each heart you spend moves the tutor one level closer to the details, up to near-pseudocode at level 5. You get only as much help as you need.

Scaffolding (Wood, Bruner & Ross, 1976)

Hearts spent → hint level

Nudge

Strategy

Step by step

Outline

Pseudocode

Level = hearts spent so far (max 5)

Out of hearts? You can still work.

Writing code and running tests never cost hearts. While you wait for the next heart, PACE suggests what to try on your own, so the pause becomes time to think it through.

Hearts throttling adapted from CS50 (Liu et al., 2024)

Next heart in 07:15

Out of hearts for now

A short break often helps more than another hint. Meanwhile:

Re-read the instructions and examples

Run your tests and compare the output

Revisit the tutor’s last hint

Coding and running tests: always free

For instructors

Instructors set the rules.

Teachers create the assignments, decide how much help students get, and review the results in one place.

Test 1 · visible

Passed

Test 2 · visible

Passed

2 hidden tests

Runs on submit

Assignments with hidden tests

Write the instructions, add visible tests for practice and hidden tests for grading, and set a due date. Scores are computed automatically.

Hearts per student

One heart returns every

30 min

You decide how much help

Set the number of hearts per student and how long each one takes to come back. The defaults are 5 hearts and 30 minutes.

Two submissions 82% alike

Review

Large paste detected

Medium

You make the final call

Flags, not verdicts

PACE flags very similar submissions and unusual activity, such as large pastes. A flag means “take a look,” not proof of copying.

Also in PACE: XP, levels, badges, streaks, and a class leaderboard. They keep practice fun, but they are supporting features and not part of what we evaluate.

The study

What we’re measuring, and what we’re not.

Students and instructors use PACE, then rate it on a 5-point scale adapted from the Technology Acceptance Model (Davis, 1989). We compare the two groups with the Mann-Whitney U test.

Four things we ask about

5 items each · 2 versions

Perceived usefulness

“Does PACE help me practice and finish my assignments?”

PEOU

Perceived ease of use

“Are the editor, tutor, and hearts easy to learn?”

ATU

Attitude toward use

“Do I like using a tutor that has limits?”

Intention to use

“Would I keep using PACE and recommend it?”

What we don’t claim

Better grades or skills.

We don’t measure learning outcomes.

A tutor that adapts to you.

Hint levels depend only on hearts spent.

That flags prove copying.

They only prompt an instructor to review.

That hearts build self-regulation.

It’s the reason for the design, not a measured result.

respondent groups: students and programming instructors

rating items per questionnaire version

point scale, from Strongly Disagree to Strongly Agree, with Neutral in the middle

university setting, so results describe PUP Quezon City only

How we built it

Built in under a week with AI, reviewed by people.

We followed the AI-Driven Development Life Cycle (Raja SP, 2025): AI proposes, the team decides, in short work cycles called bolts.

Phase 1

Inception

AI turned our goals into requirements, user stories, and units of work. We reviewed them together in mob elaboration.

Phase 2

Construction

AI proposed the architecture, data models, code, and tests. We made the technical and design decisions.

Phase 3

Operations

AI helped deploy and configure the services, with the team checking each step.

Built withNext.jsReactMonaco editorSupabasePrismaJudge0Vercel AI SDK

The team

The people behind PACE.

Students of the Polytechnic University of the Philippines, Quezon City.

Fitzjaymar Jude D. Martin

Researcher

Jhoyce Anne Niel B. Dumlao

Researcher

Christopher M. Gabayno

Researcher

Prof. Ernesto J. Odpaga Jr.

Research adviser

For study participants

Taking part in the study?

Sign in, work through an assignment with the tutor, then answer the questionnaire. Your feedback shapes the next version of PACE.