Stop Making Flashcards by Hand. Here's What Ridvay's AI Does Instead.

Making flashcards by hand takes forever. You read a paragraph, decide what's important, figure out how to phrase a question, write the front, write the back, and repeat. For a chapter of a biology textbook, that's easily an hour of work before you've studied a single thing.

Ridvay does all of that in about ten seconds.

Here's actually how it works -- not the marketing version, the technical one.

What You Can Feed It

The AI generator accepts basically anything you'd use to study:

A topic. Type "the French Revolution" and it'll generate a full deck. Good for when you need a starting point and will refine from there.

Your notes. Paste in raw text -- lecture notes, a textbook summary, anything. The AI reads it and extracts the concepts worth testing on.

A photo. This is the one that surprises people. Take a picture of your textbook page, your handwritten notes, a whiteboard from class, or slides projected on a screen. Upload it, and Ridvay's vision model reads the image, identifies the key information, and generates cards from what it finds. Multiple photos at once work too -- so you can photograph an entire chapter spread across ten pages and process them all together.

Under the hood, each input type routes to a different prompt. For images, a vision model (Gemini) handles OCR and content extraction before anything is turned into flashcard pairs. For handwriting specifically, there's a specialized pipeline that handles messy or unclear text -- because most lecture notes aren't exactly typeset.

How the AI Decides What to Put on a Card

This is where it gets interesting. The generation isn't just "summarize this content." The model is given a specific set of rules it has to follow:

• Questions must be clear and specific, not vague
• Answers should be concise -- one to three sentences
• Cards should cover different aspects of the topic, not just repeat the same idea
• Question types should vary: definitions, concepts, comparisons, applications
• The goal is testing understanding, not just memorization

The model outputs structured JSON -- a schema is enforced so the response is always an array of front/back objects, nothing else. Temperature is set to 0.4, which keeps the output consistent and focused rather than creative. You want reliable flashcards, not interesting ones.

If the model returns malformed output, there are two automatic retries before it surfaces an error. In practice this rarely happens.

The Part That Makes It Actually Useful: Spaced Repetition

Generating the cards is step one. What happens after is what separates useful flashcard tools from forgettable ones.

Every card Ridvay generates gets initialized with the SM-2 algorithm -- the same spaced repetition system Anki uses. Each card carries three values:

• Ease factor (starts at 2.5): How easy this card is for you personally. Goes up when you nail it, down when you struggle.
• Interval: How many days until you see it again.
• Repetitions: How many times you've reviewed it successfully.

When you review a card, you rate it: Again, Hard, Good, or Easy. Miss a card and it resets -- you'll see it again tomorrow. Ace it and the interval grows: 1 day, then 6, then multiplied by your personal ease factor each time.

The result is that you review cards right before you'd forget them. You spend zero time on things you know well, and more time on things you keep getting wrong.

The AI generation feeds directly into this system. The moment you save a generated deck, every card has a next review date set to today -- they're in the queue and ready to go.

From Photo to Reviewed Card: The Full Path

Here's what actually happens when you photograph your handwritten chemistry notes:

1. You upload the image through the app
2. The image goes to Ridvay's backend, which calls Gemini Vision
3. The vision model handles OCR and content extraction -- identifying formulas, reactions, concepts
4. That extracted content becomes the input to a generation prompt
5. The prompt runs through Gemini with structured output enforced
6. You get back, say, 12 flashcard pairs
7. You preview them, optionally edit any that don't look right
8. You save to a deck
9. Each card is stored with SM-2 defaults initialized
10. The deck is live in your study queue

The whole thing takes about 10 seconds for a single image. Multiple images are processed collectively, so the cards cover the full content rather than just one page.

Why This Actually Matters

The hardest part of using flashcards isn't reviewing them. It's making them. Most people who intend to use spaced repetition give up because the upfront investment of building a deck is too high.

If you can photograph your notes and have a full deck ready in seconds, that barrier disappears. You can make a deck for every lecture, every chapter -- not just the ones you had time to manually create cards for.

The AI won't always get it perfect. Sometimes a card will be too broad, or phrase something in a way that doesn't match how you think about the topic. But editing one card takes five seconds. Even editing three is faster than building the deck from scratch.

That tradeoff is worth it. And the more you use it, the better you get at spotting which cards need a quick tweak versus which ones are good to go.

Give it a try with your next set of notes. You might be surprised how quickly it becomes the first thing you do after a lecture.