Case file 05 · Tooling · Product
IDC Revision
Turns my Obsidian study vault into an exam-revision site: a dictionary for each subject, worked problem sets, and past papers you sit in the browser with done, wrong and skipped tracking. My whole class used it, with 500+ visitors in 30 days.
- Year
- 2026
- Role
- Solo: design, build, content pipeline
- Read
- 4 min
Context
An exam-revision site built out of my own Obsidian study vault. Six subjects, each with a dictionary of linked concepts, worked problem sets, annotated and explained lecture notes, and past papers you sit in the browser. I built it around the way I already revised (do a past paper, go back to the notes when stuck) and then shared it with my class.
The problem
When you're revising for exams, there's a friction with the university's student portal (Moodle) that everyone feels and nobody knows how to fix. To revise a topic you find the tile the lecture was posted under, then go into a different tile for the problem set, and by the time you have what you wanted you've lost five minutes and downloaded three PDFs you didn't need.
All year I'd been putting every lecture deck and homework assignment into an organised Obsidian vault (a folder of linked notes). So the material was all there: my notes, PDFs of problem sets, and whatever past papers each department had or hadn't handed out. But even in the vault it was hard to move around and find what I needed. Doing a paper meant the questions in one window and a hunt through the vault in another, with no record of which ones I'd already got wrong.
I thought I had a good, clear system, and I wanted to share it with my class in a form they could actually use. A vault isn't that. It's a folder of files laid out for Obsidian, and its links and layout don't work on the web as they are.
What I built
An Astro site built from the vault by an import step. The vault stays where I write and load the course material, and the site works from its own copy, so a page can be checked and tidied for the web without touching my notes. Every subject gets a dictionary built from the vault's own links between notes, problem sets with worked solutions, recipes for the procedures worth memorising, and past papers as real pages instead of PDFs, with worked solutions added where a paper came without them. Equations are typeset properly with KaTeX, so the maths doesn't look like gibberish. Where a course gave out no past papers, I generated mock ones from the course material.
The most recent addition is the graphs. Most of the static chart images, which were generated in Python, have been replaced with graphs that build up one step at a time, with a play button and a short explanation at each step. You can also drag points on a graph to change the numbers and watch it respond.
Decisions & iterations
Past papers live on the site, not as attached PDFs, so a question can link straight to the concept behind it and a wrong answer gets recorded instead of forgotten. The import only goes one way on purpose. I write in the vault and the site builds from its copy, because a website can't be built reliably straight off a folder laid out for Obsidian. Pages are generated as drafts, marked as drafts, and then reviewed, and re-running the import never overwrites a page that's already there.
The most interesting fix was in the question banks. In one sample paper a teacher gave us, the correct answer was the longest option 25 times out of 25, so it could be cracked without reading a single question. So I generated new mock papers from the material, with an automated check that the answer isn't given away by its length. The same check caught one of my own drafts, where the answer key had come out as a repeating four-letter cycle, and I reshuffled it before it went up.
Outcome
Deployed and used by my whole class, with over 500 visitors in 30 days according to Vercel Analytics. It grew from a 31-page first version, with one hand-written econometrics sample, to 745 pages across six subjects, checked by 114 automated tests. Past papers now include checked multiple-choice sets, five mock papers for the machine-learning exam alone, generated from the source material, and full worked walkthroughs, including a plain-English version of the same material for someone who doesn't code. I built it as the tool I'd revise with, and it became the one my class revised with too.



