Study Bot
A study planner that runs the kind of session research says works. That’s rarely the kind that feels productive.

The numbers
Clips


Why I built it
Generic flashcard apps treat every topic, and every student, the same. They don’t know which concepts you’re shaky on, when you’re free, or what your professor spent 20 minutes on last week. I wanted a planner that knew all three.
What I decided
- Five modes, each based on the research. They’re retrieval, interleaved practice, exam simulation (answer everything, then score it), error repair and worked examples. The repository’s learning science notes cite 44 references. They explain the research behind each principle, how the app uses it and where it stops helping.
- Rate your confidence before you see the answer. Before anything is revealed, you say how sure you are: guessing, unsure or sure. A wrong answer marked sure gets a confident error note. Feedback fixes mistakes like that well at first, but they tend to return later. That’s why they come first in error repair. Like every miss, they stay open until you get them right on two different days. The end screen shows where your confidence and your accuracy didn’t match.
- No peeking. No hint, excerpt or answer reaches the screen until you’ve answered. The server enforces that, and an end-to-end test checks it.
- Put the plan on your calendar. The week’s plan downloads as a calendar file or goes straight to Google Calendar.
- Every step of answering works from the keyboard. My goal for the redesign was level AA of the Web Content Accessibility Guidelines (WCAG) 2.2. The browser tests run axe, an automated accessibility checker, against the guidelines’ A and AA rules.
Screens



The October redesign
In July, Study Bot’s home screen was a dark dashboard: session cards, a line of stats, achievements and a year of activity squares. I redesigned the app from 11 to 16 October, in 135 commits.
On the first day, 11 October, I worked on the old app. I moved it from Next.js 14, which had unpatched critical security advisories, to Next.js 16, tightened its security and fixed bugs that had been there since July. Connecting Google Calendar always ended on a page-not-found error. Plans used the server’s clock, so a student in New York could get sessions four hours early.
Before I rebuilt Home, I compared five mockups of it.





On 12 October I built the design system before touching a screen. The colors come from Material 3, Google’s design system, and study sessions get a palette of their own. The type is Google Sans Flex. A page that only runs in development shows every color pair with its contrast measured live.
On 15 October I rebuilt the screens. A session now works like a Duolingo lesson, with one question per screen and a sheet that slides up after each answer. The new session screens also fixed a bug from July: feedback on a session’s last answer never showed, because the end screen replaced the question as soon as the server marked the session done.


Status
Study Bot runs locally with a demo account, and there’s no public build yet. The session engine, the modes, the search over your files and the planner have unit, integration and browser tests.
Rate it before you check
Try three questions from the app’s demo course. Pick an answer, then say how sure you are before you check it.
Question 1 of 3CSS specificity
A link sits inside <nav id="main"> and has the class active. Which rule sets its color: #main a { color: navy } or .nav .links a.active { color: red }?
Correct answer:
Confident error
You marked this Sure. Errors made with high confidence are often corrected well right after feedback, then return later, so this question comes back on another day.
Why your answer is tempting
Checking an answer needs JavaScript.
Question 2 of 3Box model
A div has width: 300px, padding: 20px, a 2px border and the default box-sizing. How wide is it on screen?
Correct answer:
Confident error
You marked this Sure. Errors made with high confidence are often corrected well right after feedback, then return later, so this question comes back on another day.
Why your answer is tempting
Checking an answer needs JavaScript.
Question 3 of 3CSS specificity
What is the specificity of :where(#hero) .button?
Correct answer:
Confident error
You marked this Sure. Errors made with high confidence are often corrected well right after feedback, then return later, so this question comes back on another day.
Why your answer is tempting
Checking an answer needs JavaScript.
Confidence
- Guessing:
- Unsure:
- Sure:
Error repair
Your next Error repair session asks these in this order. Misses you marked Sure come first.
No misses yet. A wrong answer lands here.
Each one stays until you get it right on two different days.
This runs in your browser. The questions and the wording come from the app, and the list orders misses the way Error repair does. In the real app the server grades your answer, and a miss also gets a repair question at the end of the same session.