Build a Telegram vocabulary review bot: flashcards, self-ratings, and progress

Looking up a word and recalling it later are different tasks. A Telegram vocabulary tool should distinguish adding study material from completing a review: the latter records what you could recall before seeing the answer.

A historical Python aiogram deployment brief requests maintenance of an existing English–Uzbek translation bot, with a personal vocabulary list, quiz mode, and word of the day as optional additions. Checked November 7, 2026, it was Closed; its absolute publication date was unknown. This article adapts the vocabulary-list idea into a proposed new private-chat tool. It is neither a delivered client case nor evidence that BotFatherV2 imports Python projects.

Ask one clear question per card

Start with manually entered prompts and reference answers. A prompt might ask for the English word for an appointment; the answer could include “appointment” and a short example sentence. The user checks the material. Automatic dictionary lookup and generated examples are separate features.

One word can have several meanings. Recognizing an English word and producing it from another language are also different questions. Store the prompt language, answer language, and context. If reverse-direction practice is added, give it a separate card and progress record rather than treating one recognition score as mastery in both directions.

Keep a stable card ID, owner, prompt, answer, content version, and enabled status. On repeated input, offer to keep, edit, or add a distinct context instead of silently overwriting an existing card. Avoid importing decks whose reuse rights are unclear.

Reveal first, then allow a self-rating

Anki's official Studying manual describes showing a question, revealing its answer, and then rating recall. It also allows a simplified Again/Good choice. The design below borrows that interaction principle; it does not implement or claim compatibility with Anki's scheduling algorithms.

Use three stages:

1. Recall: show only the prompt, with Reveal and Skip actions.
2. Compare: show the saved reference answer, then offer Could Not Recall and Recalled It.
3. Save: acknowledge a persisted rating, show the next review time, and continue.

Revealing an answer is not a successful review. Skipping, quitting, and timing out are not correct answers either. A self-rating reports the learner's judgment; it is not an automatic language assessment. The first version does not need model grading of synonyms, spelling variants, or open-ended responses.

Link each rating to a particular attempt and card version

Associate the learner, card ID, content version, attempt ID, reveal state, rating, and rating time. Keep learning progress separate from message-delivery status.

Double-clicking a rating must not advance progress twice. An old message must not grade a newer attempt, and a failed save must not produce a success confirmation. The callback safety guide covers general permission and retry handling. The learning-specific requirement is identifying exactly which version was reviewed.

If the answer changes during a review, a useful policy ends the old attempt and shows the new version. Do not attach a rating for an old question to a revised answer. Let the user explicitly choose whether editing a prompt or answer resets progress. Suspending a card removes it from future queues without necessarily deleting its history.

Query due cards before adding reminders

On /review, select enabled cards that are due and offer a small session the user can finish. Give new and not-yet-due cards explicit entry points. An empty due queue does not mean every word has been learned.

For a testable prototype, an illustrative rule could schedule Could Not Recall for 24 hours after a saved rating and Recalled It for 72 hours afterwards. These are example product rules, not validated optimal learning intervals, FSRS, or Anki's algorithm. Measure from the persisted rating time. Revealing without rating leaves the previous due time unchanged.

Returning after several missed days should produce one due card, rather than a separate debt for every missed date. Report new cards, saved review attempts, and self-ratings separately; message volume is not evidence of improved memory.

Saving a due time does not provide automatic delivery. This first version queries only when the learner requests a review. Word-of-the-day delivery, proactive reminders, and background scheduling need separate verification; see the scheduled notification guide. If you need established adaptive review scheduling, consider using Anki before commissioning a custom Telegram interface.

Where BotFatherV2 fits

As checked November 7, 2026, BotFatherV2 supports creating and modifying Telegram bots through AI conversation. The requested /en-US page was inaccessible; capabilities were verified through its accessible English homepage and Build with AI documentation.

User-isolated persistent tools are among its documented initial directions, making manual cards and on-demand review reasonable prototype requirements. This is not a ready-made flashcard product or a learning-outcome guarantee. Verify current SDK storage, time handling, and protection against repeated ratings, then test real Telegram interactions.

The fixed SDK does not support arbitrary npm packages or Node/Bun APIs, and the initial release does not add paid-AI-dependent runtime capabilities. Anki synchronization, FSRS packages, automatic translation, speech, and model grading are not default features. AI-assisted development and runtime AI are separate capabilities.

The client's existing Python/aiogram code requires deployment in its own stack or a separately scoped rewrite. BotFatherV2 imports existing Telegram Serverless bots. It is independent of Telegram's official BotFather.

A first-version brief and acceptance checks

Create a private-chat vocabulary review bot. Users manually enter prompts, reference answers, and language directions. Cards have stable IDs and content versions; all data is isolated by user. /review selects up to five enabled due cards. Show the prompt before revealing the answer, and allow self-rating only after reveal. Reveal, Skip, and Quit do not increase scores. Save at most one rating per attempt; reject ratings for old versions or attempts. Example intervals are 24 or 72 hours after saving the rating, explicitly described as illustrative. Query on demand only: no proactive notifications, translation, speech, model grading, or Anki synchronization. Report unsupported SDK requirements before generating a small prototype.

Test with invented material before launch:

• Two users cannot read, edit, or rate each other's cards.
• Reveal, Skip, and Quit leave rating counts unchanged; reopening permits review.
• Duplicate ratings save once, and buttons from a superseded answer are rejected.
• Queries immediately before and after the 24/72-hour boundaries match the displayed next review time.
• Suspended cards disappear from review; restoration follows the published progress policy.
• Records survive restart or publishing, and statistics reconcile with individual saved attempts.

These are proposed acceptance conditions, not results from a bot deployed for this article. Establish auditable review behavior before expanding the deck or algorithm.

Sources and dates

• Freelancer, Deploy Python aiogram Telegram Bot to Render (24/7), English client project 40769381. Checked November 7, 2026: Closed; absolute publication date unknown. Only the client brief informs the scenario, not bidder promises.
• Anki, Studying, official English manual without a single publication date; checked November 7, 2026. The former studying.html entry redirects here. Used for reveal and self-rating behavior, not as validation of a custom bot's algorithm or effectiveness.
• BitBear Studio, product homepage and Build with AI documentation, capabilities checked November 7, 2026 through the accessible /en pages.