Turning 138 cryptic clues into my personal learning dashboard

The app tells me if I won today. Not if I'm getting better.

Minute Cryptic is an online cryptic crossword site that publishes one cryptic clue a day. It was co-founded by Angas Tiernan and Liam Runnalls, (Liam also writes for The Age). Each clue comes with hints you can use, and a par: the number of hints and revealed letters the community typically needs.

The app tells me my streak, my time for that clue, and whether I beat today's par. It doesn't tell me whether I'm getting faster, which kinds of clues confuse me, or how I compare with everyone else.

A linguist, a dad and a five-letter minimum

I love crosswords and puzzles, and they're some of my fondest childhood memories. I think my dad is a trivia and crossword genius; at one point I was convinced he had unlimited knowledge. I grew up in Melbourne, where we had The Age. I remember flicking through the Green Guide and highlighting what I wanted Dad to record, and doing the daily Age quiz together over dinner.

I was lucky to have the kind of family where game and puzzle nights were a regular thing. We played Word Factory (a bit like Boggle), and unfortunately for me, being eight years old didn't matter. Most of the time I played at the same level as everyone else, with a five-letter minimum.

The funny thing was that Dad, despite his endless knowledge, never did cryptic crosswords. When I asked him why, he said they were too difficult and that they needed a different, more abstract kind of brain work. Of course, I was immediately curious. I saw a golden opportunity to be better than everyone else at something.

So I tried one, and... failed epically.

Many years later, as an adult, I discovered Minute Cryptic. To my delight, it was run by Australians, one of them even an Age cryptic crossword writer, with features from other Age writers!

Cryptic clues are a language puzzle. Every clue hides a definition plus some wordplay: an anagram, a hidden word, a container, a charade. As a linguist-turned-designer, I wanted to see which of those tricks my brain handles well, and which ones still slow me down.

138 clues, and no intention of typing them in by hand

I had 138 past clues and no intention of typing them in by hand. Because the minute cryptic app syncs to my membership, I figured I could compile my own solve history from my account instead. AI-assisted workflow to the rescue.

I had some experience inspecting data sources, but I didn't know where to look or how to export anything useful. So I turned to Claude, and it became my first AI project.

Finding the data

Finding the data was the hardest part. I opened my browser's developer tools, screenshotted what I could see in the console and network requests, and sent it to Claude. Claude pointed me to the right requests and explained what each one was, until we worked out that my clue solving history lived in the app's hosted database (Supabase).

One of the console screenshots I sent Claude, asking where my data actually lived

Sketching & iterating Claude's initial design

The first version of the design had no filters and an ugly green and red colour combo. I understood that Claude intended for it be easily recognisable and green is good and red is bad, but it was displeasing.

I often sent sketched lo-fi to mid-fi diagrams to speed up the feedback iterations, paired with specific prompts, like moving the time filter out of the top-right corner and into the chart header for a more minimal look.

Rather than describing layouts in words, I worked them out on a Figma board first: the page structure, the clue pop-up and the findings callouts. A screenshot of a sketch said more than a paragraph of prompt ever could.

My Figma board of mid-fi sketches, which I screenshotted to prompt Claude

Colour, legend and time filters

I swapped the original green palette for a brighter blue, orange and pink, turned each legend swatch into its own checkbox, and put the time filters (all time, 3 months, 1 month, 2 weeks, 1 week) right next to each chart title. Fewer borders, fewer controls, same power.

Insights, not just charts

Charts alone left me guessing what they meant. Each panel now has a short “What this means” note, and there's a new breakdown by clue technique. Every insight is written from the actual data, so it never claims more than the numbers show.

Sorting the solve log

I first tried a row of sort pills above the table. They added clutter, so I moved sorting into the column headers: clue date, solved, time and score, with simple arrows and clue date as the default.

Protecting paid content

Minute Cryptic's archive is part of its paid membership, and I didn't want my portfolio to give it away. The public demo shows only the two most recent clues in full; older clues are blurred, and hints and answers are removed entirely.

My first AI-assisted project!!

The primary aims of this project was to learn about how AI can be embedded into my workflow. I aimed to explore the extent of Claude's capabilities, and see how it would help ideate solutions when given constraints and goals. This was pure fun, starting as a question about my own daily habits and turned into a small end-to-end product: data, visual design, copy and front end, all in an afternoon.

My main takeaway was being impressed with how Claude can read and interpret bulk instructions, and address every problem. If this was a professional project, I imagine my design flow would look very similar to this:

1. 01Write down the real constraints, pain points, human goals and the gap from A to B
2. 02Create first mock-ups with Figma Make or Claude Design
3. 03Hand it to developers, instead of me scrambling through the inspect terminal
4. 04Demonstrate the functionality with placeholder data
5. 05Developers start the build while I improve the UI alongside

It's really good for lowering some of the initial development barriers I otherwise would've hesitated on, such as "how do I get the data from the website"

Finally, Claude assisted me as a designer with little coding background, to look out for certain design structures & & choices to acknowledge how things are actually built, such as considering optimisation problems. This saves a developer from having to answer a million questions from me.

Keep the learning curve growing

Right now the dashboard is a snapshot from 30 September 2026. Next, I'd like a small userscript that updates it after each solve, so the learning curve keeps growing without any manual steps.