StudyDeck.io - building the study tool I needed at 30%
A case study on shipping AI to people I love (me first!), and some product tips for other builders
TLDR;
StudyDeck wasn't a startup idea. It was a survival tool that quietly turned into something I'm proud of, mostly because the people I built it for keep telling me it works.
If you're reading this, the takeaway I'd offer is simple: the most defensible products tend to come from you solving a problem you have personally bled for, with the discipline to keep saying no to everything that isn't the point.
I built this first for me. Now, perhaps for you?
You can just do things...
I was working a full-time job on a tight schedule when I enrolled in aviation school for evening classes. I joined the semester 6 weeks late (very whimsical decision - I was trying to get over a broken engagement lol). By the time I'd caught my breath, the pre-CAT was already on me, and I'd been to maybe a handful of lectures.
Naturally, I flunked it. 30% in Human Performance which is one of the easier, more enjoyable subjects for me (damn good instructor). So I'll not get into what I scored in the rest. Of course, this was a knock against my Type A ego.
The Decision Making Distance was closing in on me... I had 4 weeks to get through 10 weeks worth of class. And my full-time job at a high-growth startup. Yes, my Concerta does some heavy lifting, but rereading copious notes across 7 units wasn't going to do it for me; I needed to know what I didn't know - and then drill it in until I did.
I had done all the back and forth with ChatGPT from my notes and I needed to find out, "will I be kicking a** at these tests or is mine about to get kicked (!)" So one frustrated evening, I created my own exam-prep tool to gut check my understanding of concepts that up until 4 weeks prior, was greek to me.
That tool became StudyDeck.
My HP score came back as 70% (see pic at the bottom) in the main CATs that I did 2 weeks later. Right at the passmark for my PPL licensure requirements per unit.
This number mattered, yes, but what I remember more clearly is the feeling of sitting down already feeling prepared. Nothing like it.
StudyDeck turns your own notes (lecture PDFs, docx, etc), or the world wide web- into quizzes you can actually trust before an important exam. It's built for the moment a few days before the test, when you're past reading and you need to know what you still don't know.
Try it: studydeck.io.
What it is, short version:
When building is cheap... your human judgement + expertise + creativity become the secret sauce.
The barrier to building stuff has effectively plummeted, and if you're online enough you'll notice the results of this: a lot of what folks now call 'AI slop'. ie a product without soul, just a bunch of features existing next to each other, but not really dancing together.
There are a hundred AI study tools now. A lot of them appear over-engineered, some had sleek UIs but they weren't built to help me feel prepared for an exam. Also, I wasn't sure whether they would handle the kind of work I needed done with my technical notes and the trap questions we get in aviation exams.
Remember, the goal is not the app. The goal is passing exams. So, I needed to figure out how to digitize a prep strategy that will get me to the end goal.
I pored through research on memory and study. And ended with a deliberate bet on two ideas supported by decades of cognitive-science evidence behind them:
-
Active recall: testing yourself moves a fact from "I've seen this" to "I know this." Re-reading does not.
-
Spaced repetition: bringing missed questions back tomorrow, then in three days, then in a week, right before you'd forget.
Everything in the product serves those two mechanisms. The AI [and how I fine-tuned the LLM to know how to generate the exact right type of questions] is the means; the recall loop is the product.
Every new darling idea I had; eg, a summary view, or an explainer mode, I kept asking: does this make the user do recall, or does it let them avoid it? Brutal, but effective guardrail, I found. I can be a runaway builder with these things.
Advice for other builders: Features =/= impact. Keep the true end goal in mind; and use that to filter what stays in or out. The self-audit is brave, and efficient.
The superpower that is solving a problem you are intimately familiar with:
Many apps are designed for what their creators consider the user's happy path.
I designed StudyDeck for an emotional state I could relate with; the quiet panic at 11pm three days before an exam.
A few decisions that came out of that:
-
Exclusively dark mode: Eyes are tired. But also, which student uses light mode nowadays- lol.
-
No leaderboards, no un-earned confetti: In the exam-prep stage, you're competing with yourself and the goal is to walk in feeling prepared so you can now compete with others. Anything that doesn't feed into this is unnecessary, esp early on.
-
A mascot named Decky who speaks rarely, and never cheerleads. Decky is a steady presence, not a hype man.
-
Plain language everywhere. "Generating your deck" instead of "AI is processing your request."
-
Buttons that say what will happen so no bandwidth needed to decode fancy jargon (and accessibility tools can read them easily).
Tone, I insist, is the part of a product people feel before they can articulate. Get it wrong and no amount of features fixes the bounce rate from folks who'll feel they're in the wrong place- even if they cannot put a finger on why.
Advice for other builders: Get as close as possible to the problem you're hoping to solve.
And let your insights show up in the tonal experience of your product.
The dogfood is served... who's eating?
The first three (non-Cindy) users were the people closest to me, and each was prepping for an exam where the cost of failure was a year, a license, or a career:
-
My elder sister, Dr. Adem-Amenge: a pediatric gastroenterologist preparing for her Medical Fellowship exams. She passed, and graduated as the only locally-trained gastro-ped in Kenya.
-
My younger sister, Eng. Adem: a graduate civil engineer preparing for her professional certification (and most lethal critic)
-
My best friend, Cesy: going through nursing school one large text book at a time (she wanted the app named after her but, oh well)
They were generous, and brutal in the way only family can be.
The most useful piece of feedback I got came from my civil-engineer sister: she uploaded a 300 page manual on Geometric Design of Highways, Rural Roads, and Urban Roads- a hefty engineering document. She typed a generic topic, and got questions she described as "too light, I was hoping for the hard stuff."
Interesting. And it made sense- the model had nothing sharp to anchor on, so it generated the safest possible questions.
Her feedback inspired previously underestimated UX upgrades: gentle nudges in the UI that teach users how to phrase a topic so the AI has a more targeted cue to bite into. Not a "type more"- type sharper. "Separation" beats "Air Law" every time if we want to prepare for specific skill areas. And, 'separation' in aviation is different from 'separation' in say, Marriage Law. So, also encouraging users to add context (not just topic) helps us help them.
And yes, I also talked to the model to sharpen how the system would handle the processing of uploads with hundreds of pages.
That single piece of feedback was worth more than three weeks of analytics.
The people who love you enough to tell you when something is mid will save you from shipping mediocre, use them.
Advice for other builders: Your earliest users should be people who can tell you the truth without softening it. Then make it easy for them. Check out the Mom Test which is a classic on this topic.
What I left out, on purpose:
Here's my religion: Just because we can, doesn't always mean that we should.
Building is fun and mad addictive. But, every "nice to have" is a tax on the things that matter.
Things I deliberately did not build, but might be convinced to in future if I see it solves the emergent needs of our users:
-
Public profiles, leaderboards, social feeds. Feel too busy and can take focus away from the actual task at hand; exam prep.
-
A general-purpose chatbot. ChatGPT has that covered for now.
-
Streak counters, daily reminders, gamification of any kind. This was a hard one to leave out; I love a good celebration. Might bring it in a bit later.
-
A "summary" view of uploaded notes. It would have been easy to ship and would have actively undermined the whole thesis of the product.
For now, I didn't want StudyDeck to feel like another app pulling at your attention. I wanted it to feel like the desk lamp you turn on when it's time to work.
Advice for other builders: The features you don't build define the product more than the ones you do. Write your "not now", "not ever" list as you plan out your 'must haves'. Remember, just because you can, doesn't mean you should. And you can always change your mind later.
Where next for StudyDeck?
PostHog tells me as at now (30th April, we go live officially later today), I have 100 visitors, and 3 people who are not my friends nor family who have created accounts; ain't much but it's honest work (and thanks Reddit!).
The roadmap is honest and short:
-
Installable on your phone: A proper home-screen experience for iOS and Android, because making exam prep mobile-first just makes it easy for users to carry it wherever they go.
-
Interview-prep options: Small sis (engineer) told me she wants to prepare for her oral interviews before the Eng. Certification Board. Is there something there, perhaps? We'll see.
-
Calculation-style questions. Esp for technical notes with potential trick questions. Lots to consider here when it comes to the UX a user studying an arithmetic question would find most useful.
Each of those is a real user request, traced back to a real moment in someone's revision week and the feedback they offered to me.
My name is Cindy; I build here at product-queen.com.
If you're a builder who has perfected the art of concept-to-market by leveraging AI as much as possible, I'd love to chat! I always respond to all feedback channeled through the chat icon on the bottom right. And comments left at the bottom.
Oh, and here's the receipts, as promised:
Let’s take this offline?
Working on something we can collaborate on? Write to me- I'll respond.