Photograph the worksheet, get it explained in the language you picked once, turn the terms into flashcards, and let the agent run the reviews. Written from the questions a student actually asks.
Fasrad · September 16, 2026 · 4 min read
"Here's tonight's worksheet. Pull out the questions and tell me what each one is really asking."
That's the message Maya sends most evenings, with a photo of a handwritten page attached. She's a junior in Portland taking AP US History and Spanish 3. A Fasrad agent reads the photo, works from her own material rather than a textbook, and from that text it explains the material or turns it into flashcards, in whichever language she set. Then it reminds her to review them.
Yes. Drop the photo in and ask for what you need. The agent reads the text off the image and keeps working from it for the rest of the conversation.
For that worksheet it came back with the four questions numbered, each with a one-line reading of what the teacher is actually after, and then it waited. Photograph the page flat and in decent light. If a line came out wrong, say which one it was and paste the correction. It carries on from the fixed text.
"For anything about Spanish class, answer in Spanish first and put the English underneath." Said once, that becomes a standing instruction, and every later conversation follows it on the web and on Telegram. Changing it takes another sentence whenever the teacher changes the rules.
The same memory holds smaller preferences. Maya asked for a retelling of a short story "in Spanish, at a level a second-year student could follow", and the agent now defaults to that register for her Spanish 3 reading. An AI language tutor built on the same agent keeps the vocabulary from those retellings for the flashcards below.
Yes, and it starts from her words instead of a textbook glossary. After a worksheet has been read, "Make a deck from these notes, term on the front, meaning on the back" pulls candidate terms out of the text and shows the draft list first. She strikes the two she already knows cold and confirms. The deck exists. A photo of a handwritten list, a pasted paragraph or a CSV export all go through the same preview.
The Reconstruction deck came out at 24 cards. Each one carries the term, the meaning in her own phrasing, and the sentence from her notes it came from. Reviews run on spaced repetition, so a card she gets right comes back at a widening interval and a card she misses returns sooner. Cards she keeps getting right graduate from recognition to fill-in-the-blank and then to producing the term herself. Every correct answer earns points, nine for a recognition card and up to eighteen for production, and a streak counter keeps the days honest.
Two ways, and the first one needs no setup. The agent watches the deck on its own. If she usually studies around 4:30 and hasn't started, if five or more cards are due and untouched for eighteen hours, or if a streak is about to break after 8 PM, it sends one short nudge.
She can also be explicit about it: "Every weekday at 4:30, quiz me on whatever is due."
The quiz runs in the chat, one card at a time, so Telegram works exactly like the web. She types an answer, the agent grades it and allows for a typo, then moves to the next card. Six cards take about three minutes. Typing "stop" ends the session. On Sunday a short digest lists what she reviewed and which cards keep slipping, and "Make me a PDF of my most-missed terms" gets her a one-page sheet for the bus. That's the everyday shape of a Telegram AI assistant.
Ask for the register you want and it stays there. Maya's version runs "Explain the Compromise of 1877 in my own words, not too formal, and then give me a three-sentence version." She gets the long one for understanding and the short one for the margin of her notes. She asked for both twice, so now it offers both without being asked.
Follow-ups stay in the thread. "Why did the North go along with it?" gets an answer that refers back to the worksheet she uploaded an hour earlier rather than a fresh lecture. When she wants to test herself, "Ask me two questions about this and grade my answers" turns the explanation into a short oral exam.
An AI tutor that remembers the unit is a different thing from a search result.
Same agent, same chat. "Remind me every two hours today until 8 to clean my riding gear" creates a recurring reminder with an end time, delivered on Telegram, and it stops on its own. "Remind me Thursday at 7 to bring the permission slip" is a one-off.
Nothing about that requires a second app. The quizzes, the explanations and the riding gear all live in one conversation.
The explanations get written from her notes and her questions, and the flashcards are her terms in her phrasing. Ask "What did you base that on?" and the agent points at the lines of the worksheet it used.
Her study material is private to her account, the Study skill stays off until she turns it on, and nobody else sees her decks.
Yes. Photos of handwritten pages, typed notes, pasted text, and uploaded PDFs are all read the same way. For flashcards, a photo of a list, a pasted paragraph, or a CSV file each go through the same preview step before cards are saved.
Anything with terms and meanings: history, biology, vocabulary in any language, formulas with their names. Cards hold a front, a back, a context sentence from your notes, and an optional memory hook, and the scheduling works the same across subjects.
Yes. The quiz runs inside the conversation, one card at a time, so it works on Telegram exactly as on the web. You type the answer, the agent grades it and moves on. Typing stop ends the session.
With spaced repetition. A card you get right returns at a longer interval, a card you miss returns sooner, and cards you keep getting right move from recognition to fill-in-the-blank to producing the term yourself.
Yes. It outlines from your notes, works through a draft with you paragraph by paragraph, tells you what is unclear and why, and checks a finished draft against the assignment prompt before you submit it.
No. The study data is private to the student's account, the Study skill is off until the student enables it, and the agent never shares decks or conversations with anyone.