skip to content

Privacy Policy

Last updated: 2026-11-17

1. Who we are

KataJob is an interview-preparation tool. You can explore roadmaps, read questions, take quizzes, and track spaced-repetition review without an account; creating an account adds sign-in, email verification, optional two-factor authentication, and progress that follows you to another device.

A registered account can also answer a question — out loud, or by typing or dictating it — on a review card, on a question's page or in a learning session, and have the answer evaluated by AI, and can ask an AI about a question in a chat in each of those three places, or about a lesson on its pages, by typing, dictating or talking. Both are paid for with credits that you buy, redeem from a promo code or are given by us. These features are switched on gradually, so you may not see them yet; this policy describes them in full all the same.

Using KataJob without an account is not the same as leaving no trace. The first time you record progress we create a guest account for you on our servers and keep your work there — section 2 explains exactly when that happens and section 7 says how long it lasts.

This policy explains what data we hold, why we hold it, how long we keep it, and the choices you have. KataJob is a site name rather than a company — it is run by an individual based in Tbilisi, Georgia, reachable at [email protected], who is the data controller for everything described here.

2. Data we collect

Account data — when you register, we store your email address, your display name, and a securely hashed password (we never store the password itself).

Sign-in with a provider — you can sign in with Google, GitHub or LinkedIn instead of a password. From any of the three we take two things: the email address the provider vouches for, and a stable identifier for your account at that provider so we recognise you next time. We never receive your password at the provider, and we do not keep the access token the provider issues. Your display name is taken from the part of your address before the @ sign; we do not copy a profile name or a profile picture from any provider. Google and LinkedIn state whether an address is verified as part of the sign-in response. GitHub does not, so for GitHub only we make one further call to its /user/emails endpoint to find an address it has verified — that is the single case where we read more than the sign-in response itself returns, and we keep only the address it gives us.

Authentication & security data — to keep you signed in and your account safe, we process short-lived JWT access tokens, refresh tokens stored in HttpOnly cookies, one record per active session (the browser user-agent string, the IP address the session was created from, and timestamps), and, if you enable two-factor authentication, an encrypted TOTP secret plus single-use recovery codes stored as hashes. We also keep failed-sign-in counters to support temporary account lockout that protects against password-guessing.

Learning data — the learning items you create (the roadmaps you are working through), your per-question progress, your spaced-repetition review schedule, your quiz history, your progress through lessons (the pages you opened, your answers to their questions, the quiz runs you finished, and whether you reached the end), the lessons you added to your own list, and any answers you write out yourself. All of it lives on our servers, whether or not you have signed up. Using KataJob without an account creates a guest account: the first time you do something that records progress — grading a card, saving a quiz, starting a roadmap, writing out an answer, answering a question in a lesson — we create an account for you and key that work to it. Reading and browsing on their own create nothing, with one exception: a lesson you read through to its last page, by its own buttons, is recorded as completed. Signing up later does not move or merge anything: it attaches your email and password, or a provider identity, to the account you already have, so the work you did as a guest is already on the account you just created.

Preferences — settings such as your reading width, your dashboard layout, your review timers and your keyboard shortcuts are stored on your account. Your browser also keeps a copy of some of them, along with your light/dark theme choice, so a page can render correctly before your account has loaded. That copy is a cache of settings, and clearing it loses nothing. A few choices are kept only on your device and never sent to us: where you last left the chat's button, and which voice you chose to talk to the chat in; clearing your browser's storage resets them. While you are signed in, your browser also keeps a short note of how your review, dashboard and learning pages last looked — counts and yes/no facts only, such as how many roadmaps or recent quiz results they showed, whether a chart was drawn and whether a roadmap was paused, never a question, a name, a score or a date — so the next visit can draw them at their real size before your data arrives. It is used only while you are signed in on this device, and it is deleted when you sign out, reset your learning data or delete your account.

Notifications and time zone — if you turn notifications on, we store which topics you turned on, the hour you picked and which of your channels each one goes to, and the channels you connect. For Telegram that is the id of the chat between you and our bot; we do not store your Telegram name or username. For browser notifications it is the push subscription your browser hands us — an address at its push service and the keys used to encrypt a message to it — and a short label such as "Chrome · macOS" so you can tell your devices apart. For ntfy it is the name of the topic we made for you; anyone who knows that name can read what is sent to it, so it is shown to you only while you connect it. For email we keep no second copy of your address: reminders go to the address on your account, and we store only that you confirmed it. Each message we send leaves a delivery record: the topic, the channel, when it was sent and what the channel answered. Delivery records are kept for 90 days. We also store your time zone: the zone you set in your account settings, if you set one, and the zone your browser last reported, which is used when you have not set one. The time zone decides where your day starts for reviews and when a daily reminder is sent.

Feedback and bug reports — if you send a report from inside the product we store what you wrote (its type, title and description), any file you attach, and the question or topic the report names. Filing a report needs an account, and the report is linked to it: we email the address on that account to confirm we received the report and, when we resolve it, to tell you what was done.

Evaluated answers — you can answer a question out loud, or type or dictate it. Out loud, your browser asks you for the microphone, records your answer and sends the recording to us with, where your browser can measure them, the moments you fell silent: the start and length of each silence of a second and a half or more, which your browser measures as it records. We pass the recording to OpenAI to be turned into text and never store it: it exists on our server only while the upload is handled and turned into text, then it is discarded. A typed or dictated answer reaches us as text, and before any of it is stored, OpenAI's moderation check reads it. A text the check flags is refused, without charge, and none of it is kept: the answer stays in your list marked as flagged, with its question and the time, and what remains beside it is the usage record of the check and the released hold in your credit ledger, neither of which holds any content. What we keep of an answer is the transcript as it was heard and a cleaned-up copy (for a typed or dictated answer, the text itself), whether you spoke, typed or dictated it, where you gave it, and the evaluation: the language it was in, whether it was on topic and whether it seemed to give the evaluator instructions; the key points it was checked against, each with its level, whether the score counted it, whether you covered, half covered or missed it, the words of yours that showed it and, for one half covered, what was missing; what you got wrong, and any interview red flag it matched; what is worth adding; a stronger version of your answer, with the key points each of its sentences states; for a spoken answer, how long it was, how fast you spoke over the whole take and over the time you were speaking, your filler words, restarts and pauses, where each falls in the transcript, and the silences your browser measured; and a score and a suggested grade. We also keep the grade you then gave, how long the recording was and its size, the question it answered with a fingerprint of the question as it read at the time, which key points it was checked against (the question's own, ones AI wrote for it, or its recall hints), which AI models and versions produced the evaluation, what it cost us, and the credits it cost you with your balance right after. If you report an evaluation with disagree / suggest, the report names the answer, and any re-evaluation we run on it is kept with the answer. We also keep the date you last agreed to the notice shown before an answer, and which version of it you saw.

Key points — every answer is checked against its question's key points. Where a question has none of its own yet, an answer to it has OpenAI write them from the question and its reference answer, with none of your words; we keep them with the question, for every later answer by anyone, linked to no account.

Dictation, talk and reading aloud — dictate here in the answer box and dictate in a chat use your browser's own speech recognition, and so does talk in a chat in the browser's voice: your browser, not KataJob, turns what you say into text. Depending on the browser, it may send the audio to its maker to do so — Chrome to Google, Edge to Microsoft, Safari sometimes to Apple — or keep it on your device, as Firefox does; that is the browser maker's processing, under its own terms and privacy policy, not ours. With dictation and talk in the browser's voice, we never receive the audio: only the text you send, and that it was dictated or spoken. A spoken turn is sent when you press to send it, or when it reaches its length limit (2,000 characters, or a minute of recording in the natural voice), never because you fell silent; while you talk, your browser measures how loud your microphone is, to pause when you stop speaking, and that measurement never leaves your browser. listen and read aloud hand the text on your screen to your browser's own speech synthesis, and so does talk in the browser's voice for each reply; we send nothing for them. A browser's voices may be online ones — Edge's voices named Online are — and with one of those the browser sends the text to its maker to be spoken.

Chats — if you ask the AI about a question or a lesson, we keep the conversation, one for each question and one for each lesson: each message you write or speak and the reply to it; for a lesson, the page of the lesson each message was sent from — its cover, a slide or its last page; whether you typed, dictated or spoke each message and, for a spoken one, whether your browser or OpenAI turned it into text; whether the reply was completed, cut off at the length limit or failed, and why it failed; which AI model and settings, and which version of our instructions, produced it; which of your voice answers was given to the AI with it; what the reply cost us and the credits it cost you; and when the message was sent, when its reply began and when it was finished. For a message spoken in the natural voice (below) we also keep how long its recording was and, if its reply was read aloud, whether the reading finished or failed, what it cost us, the credits it cost you and when it was asked for. A message whose reply failed stays in the conversation, marked as failed, and so does a spoken message that could not be turned into text, with no words. Before a message is answered, OpenAI's moderation check reads it. A message the check flags is refused and deleted straight away, without charge, and none of its text is kept: what remains is the usage record of the check and the released hold in your credit ledger, neither of which holds any content, and a count of the kind of content that was flagged, which names no account. If you report a reply with disagree / suggest or flag, the report names the message and, on a lesson, the lesson and the page it was sent from.

Talking in the natural voice — talk in a chat can use OpenAI's speech instead of your browser's, once you agree to the notice shown before your first such turn. Then your browser records each turn from your press until you press to send it or it reaches a minute of recording, and sends the recording to us. We pass it to OpenAI to be turned into text and never store it: it exists on our server only while the upload is handled and turned into text, then it is discarded. Its words are then checked, answered and kept as a typed message's are. The text of the reply is sent to OpenAI to be read aloud in a voice generated by AI, and the audio is passed straight on to your browser to be played; we never store it. We keep the date you last agreed to that notice, and which version of it you saw.

Credits — the AI features are paid for with credits, and each account has a ledger: every batch of credits it received (bought, from a promo code, or granted by us), how many of them are left and when they expire, and every movement — a charge, a hold while an AI action runs and its release if the action fails, an expiry, and any correction we make, with its written reason. If you redeem a promo code we keep that you did and when; if we email you one, we keep a record of that email. For every call we make to OpenAI we keep a usage record — which feature, which model, how much audio or text it processed, what it cost us, how long it took and whether it worked — with your account id and none of the content.

Purchases — you buy credits with Telegram Stars, inside Telegram. For each purchase we keep its reference, the pack, the number of Stars and credits, when the invoice was made, paid and refunded (if it ever is), your Telegram user id, the ids Telegram gives the payment, and which version of the purchase terms you accepted and when. We never see your card or bank details: you pay Telegram, under Telegram's own terms. We also keep a copy of our bot's Stars transaction history for our tax records, with every name and Telegram user id taken out of it.

Sign-in and activity records — we store the date and time you last signed in, and, separately, one record for each calendar day (counted in UTC) on which your account was active. An activity record is an account and a day, nothing else: not the pages you read, not how long you stayed, not what you answered. One is written when you sign in and each time your session is renewed as you keep using the site, for a guest account as well as a registered one. We read them to count how many accounts are in use over a day or a month and to see which accounts have gone unused. That is our own usage reporting: it is not advertising, and it is not shared.

Operational data — standard server logs needed to run and secure the service: the request path, method, status, timing, a request id, the IP address the request came from, and the browser user-agent. Where a log line names a user it uses the numeric account id, never an email address. If a page hits an error on your device we also record the error message, an opaque error code, the route it happened on, that request id, and which release of the site the server was running — never the stack trace. We do not run third-party advertising, analytics, or cross-site tracking of any kind.

Visit statistics — to know which pages are read and where visitors come from, our server counts page visits as it handles them. For each visit it records the page address without its query string (apart from any utm campaign tags the link carried, and with any account-specific id in the address replaced by a placeholder), whether the page was loaded in full or reached by moving within the site, the domain of the site that linked to the page, your country as reported by Cloudflare, your browser, operating system and device type, whether the visitor is a known search-engine or AI crawler or an automated client presenting itself as a browser, whether the request carried the standard headers modern browsers send with every request (Fetch Metadata and Client Hints: a yes or no for each, never what they contained), and a visitor code. The visitor code is computed from your IP address and browser with a key that is replaced every day and never stored, so that a visitor is counted once a day. The statistic itself holds neither your IP address nor your browser string — the ordinary server logs described above do, and are kept alongside it — and because the key for each day is thrown away, visitor codes from different days cannot be matched to each other. There is no cookie, no tracking script on the page and no third party involved. These records are part of the server logs and are kept for the same time (section 7).

Bot protection — to stop automated copying of the public question, topic and roadmap pages, our server counts how many of them each IP address (and, for IPv6, each block of addresses) reads per minute, hour and day. Those counts are kept only in the server's memory: they are never written to disk or to the logs, and they are gone at the next restart. A visitor who reads faster than a person would is shown a human check (Cloudflare Turnstile) before the next page. Only then does Cloudflare see anything beyond its usual role as our edge: to tell a person from a script, it processes technical signals from your browser, and we send it your IP address with the check's answer to confirm the result. When the server limits or checks a visitor, its log records what happened (for example that a check was passed, or that an automated client was refused), the page, and the same daily visitor code the visit statistics use, not your IP address. If an address keeps reading past the limits without passing the check, or opens a link that is hidden from people and that search engines are told not to follow, the record also names its network range (the first three parts of an IPv4 address, or the first 48 bits of an IPv6 one) so that we can block that range at Cloudflare. These records are kept with the server logs (section 7).

3. How we use your data

We use the data above only to operate the service: to authenticate you and keep your session secure, to save and sync your learning progress, to schedule reviews, to evaluate the answers you give, answer the questions you ask in a chat and keep both for you, to keep your credit balance and charge for the AI features, to sell you credits and keep the records the law requires of a sale, to send the account email and the notifications you turn on (section 5), to enforce security controls (such as lockout, rate limiting and two-factor), to triage the bug reports and feedback you send us, to measure how many people visit and which pages they read (the visit statistics in section 2), to protect the public pages from automated bulk copying (the bot protection in section 2), and to maintain and improve the platform.

For the AI features and credits, the legal bases are these. We evaluate and keep your answers, answer and keep your chats, keep your credit ledger and sell you credits because that is the service you ask us for. We keep a purchase record after you delete your account because Georgian tax law obliges us to (section 7). We send offers email only with your consent, which you can withdraw at any time (section 5). We keep the usage records of AI calls, and check each chat message and each typed or dictated answer with a moderation service before it is answered or evaluated, in our legitimate interest in paying our AI provider correctly and stopping abuse of the AI features.

To answer a support request about your account, we may view your learning progress (your dashboard, your review queue and your learning plan) as you see it. Viewing it changes nothing on your account.

We read an answer or a chat of yours only when you ask us to. If you report an evaluation with disagree / suggest, we read that answer and its evaluation, and may have it evaluated again to check the evaluator, for as long as the report is kept; deleting the report, the answer or your account ends it. If you report a chat reply with disagree / suggest or flag, we read that one message and its reply, with what it was about — the question, or the lesson and the page it was sent from — how you gave it, the model, the version of our instructions and the cost behind them and whether a voice answer was given with it, and for a message spoken in the natural voice how long its recording was and how reading its reply aloud went and cost — not the rest of the conversation, and not that voice answer — for as long as the report is kept; deleting the report, the conversation or your account ends it. Likewise, to answer a question about a payment or put a credit problem right, we may view your credit ledger and purchases and add or correct credits; every correction carries a written reason that stays in the ledger.

We do not sell, rent, or trade your personal data, and we do not use it for behavioral advertising.

4. Cookies

We use cookies that are strictly necessary to run the service. There are six, and only the first carries a sign-in credential:

  • refreshToken — secure, HttpOnly and same-site. This is what keeps you signed in, and it is the reason a guest account survives a page reload.
  • katajob_session — a flag rather than a token: it says only that a session may exist, so the app can skip a pointless sign-in check for a visitor who has none. It is readable by the page scripts, which is precisely why it carries nothing else.
  • katajob_rid — the id of the page request, kept for 15 minutes. Deliberately readable by the page scripts so an error screen can show you the id of the request that failed. It identifies a request, not a person.
  • kj_pass — set only if you pass the human check described in section 2, and kept for 24 hours (15 minutes if the check service could not be reached). It is HttpOnly and signed by our server, and it holds a random pass id, its expiry time, the number of pages it allows and a short hash of your browser's user-agent string, so that it works only in the browser that passed the check. Its one use is to let you keep reading without being checked again. It is not used for tracking or analytics, and it is not linked to your account.
  • katajob_started — the keys of the roadmaps you have started (the same names that appear in their addresses, such as java-developer), and nothing else: no account id and no progress. The page keeps it up to date while you use the site, so that when you open /roadmaps, /quizzes or /learning it can draw your roadmaps in their place straight away instead of the page jumping when they arrive, and the page of a roadmap and of its quiz can show the right button from the start. It is used only while you are signed in on this device, lasts at most 30 days from your last visit, and is sent only to katajob.com. It is deleted when you sign out, reset your learning data or delete your account, and the next time the site loads without your session — so the next person to use a shared device never sees it. It is only a layout hint: your roadmaps are always read from your account.
  • katajob_lessons — the keys of the lessons you have started, completed or added to your own list (the same names that appear in their addresses, such as cap-theorem), and nothing else: no account id, no answers and no dates. The page keeps it up to date while you use the site, so that when you open /lessons it can draw your lessons first straight away instead of the list jumping when your progress arrives. It is used only while you are signed in on this device, lasts at most 30 days from your last visit, and is sent only to katajob.com. It is deleted when you sign out, reset your learning data or delete your account, and the next time the site loads without your session. It is only a layout hint: your lesson progress is always read from your account.

Access tokens are held only in memory for the current session, not in a cookie. While access to the site is restricted, one further strictly necessary cookie (katajob_gate) records that you passed the access prompt so you are not asked again; it exists only for as long as that restriction does.

We do not set third-party advertising, analytics, or cross-site tracking cookies.

5. Email and notifications

We send account email — address verification, password reset, security notices, and the confirmation and resolution notices for reports you file. You cannot turn these off while you have an account. Everything else is a notification you choose at /account/notifications, on the channels you connect. All of them are off until you turn them on, and each can be turned off from the message itself or from that page.

Account email is delivered through our email provider, Resend, which processes the recipient address and message content solely to deliver the email on our behalf. A notification is carried by the service behind the channel you connected, named in section 6. We do not send marketing email without separate consent.

Offers — emails carrying a promo code for AI credits — are the only marketing email we send, and only if you ask for them: the box at sign-up and the switch in your notification settings are both off until you turn them on. Every offers email carries an unsubscribe link. When you buy credits, we email you a receipt and our Telegram bot confirms the purchase in the chat you paid from; both repeat what you accepted at checkout. Like account email, a receipt is sent whatever your notification settings say; Amazon SES delivers it (section 6).

Reminder email goes to the address on your account, and only after you confirm it from a link we send there. Every reminder email carries an unsubscribe link, and mail apps that show their own unsubscribe button use the same link; it stops that reminder by email and leaves your other channels alone. If a reminder email bounces permanently — the address cannot receive mail — we stop sending reminder email to you. If you report one as spam, we stop too, and take email off every reminder.

6. Service providers

We rely on a small number of processors to run the service. Naming them is more useful than naming a category, so here they are:

  • Railway — hosts the application and its database, and holds a copy of the server logs described above.
  • Cloudflare — the edge that every request to the site passes through, Email Routing for mail sent to our published addresses, and Turnstile, the human check described in section 2. Turnstile is used only when a visitor is checked.
  • Resend — delivers the account email described in section 5.
  • Telegram — delivers reminders to a chat you connect; Telegram learns that our bot messages that chat.
  • Your browser's push service — Google, Mozilla, Apple or Microsoft, depending on your browser, carries an encrypted reminder to your device and cannot read it.
  • Our own ntfy server — run by us; delivers reminders to the ntfy app if you connect it. If you use the ntfy iOS app, ntfy.sh receives a content-free signal that a message is waiting.
  • Amazon Web Services (Amazon SES) — delivers reminder email, offers email and purchase receipts, from its Frankfurt region (eu-central-1). It processes your address and the message to deliver it, tells us when a message cannot be delivered or is reported as spam, and keeps an address that could not be delivered to or reported spam on a list that stops further mail to it.
  • OpenAI — turns a spoken answer into text, checks a typed or dictated one with its moderation service, evaluates every answer, writes the key points of a question that has none, replies to the messages you write or speak in a chat about a question or a lesson, and, when you talk in the natural voice, turns what you say into text and reads the replies aloud, in the United States. For a spoken answer it receives the recording, then the transcript with the question and its reference answer. For a typed or dictated answer it receives the text on its own first, for its moderation check, and then the text with the question and its reference answer. To write a question's key points it receives the question and its reference answer, and none of your words. For a chat message it receives the message on its own first, for its moderation check, and then, to reply, the message with the question and its reference answer, your earlier messages in that chat with their replies (the most recent that fit), and your latest evaluated answer to the question if you have one — its transcript or text, its score and what the evaluation said you covered, missed, got wrong or could add. For a message in a lesson's chat it receives, to reply, the message with the lesson's title, its introduction and its outline — each slide's title and its one-line summary — the slide you sent it from in words, with its figure's captions and labels and the terms it uses (on the cover, the lesson's red flags and interview questions instead; on the last page, nothing more), that slide's quiz questions with their options and, for each one you have answered, your pick, the right option and why, the quiz questions you got wrong across the lesson with your pick and the right option, and your earlier messages in that chat with their replies (the most recent that fit). For a message you speak in the natural voice it receives the recording first, to turn it into text, then the text as for a message you write, and then the text of the reply, to read it aloud. Under OpenAI's own terms for its API, it keeps no copy of a recording and nothing from the moderation check, keeps the text it evaluates, replies to or reads aloud, and the audio it reads it in, only in abuse-monitoring logs, for up to 30 days unless the law requires it to keep them longer or it needs them longer to protect its services or anyone else from harm, and uses none of it to train its models.
  • Telegram (payments) — sells you the Stars you pay with and processes each payment, under its own terms and privacy policy. It tells us your Telegram user id and the ids of the payment, and nothing about how you paid for your Stars.
  • Google — mail forwarded by Cloudflare arrives in a Gmail inbox we read and reply from.

We also run our own monitoring server, which holds a second copy of the server logs. These providers process data only on our instructions and under appropriate agreements, except Telegram's payments, which Telegram runs under its own terms as described above. A current list is available on request at [email protected].

7. Retention

We keep your account and learning data for as long as your account is active. A session lasts one day for a normal sign-in and thirty for a "remember me" one. Revoking a session ends it at once, but its record — with the IP address and browser user-agent described in section 2 — is kept until thirty days after the session would have expired, and a daily sweep then deletes it. The record of a session that simply runs out is deleted the same way, thirty days after it expires.

Guest accounts have a time limit, because nobody is there to close them. A guest account that has gone unused for 180 days is deleted, and everything keyed to it goes with it. So is a guest account whose data a real account has already taken over, seven days after the takeover. Clearing your cookies makes a guest account unreachable to you, and the same sweep is what eventually removes it.

An account whose email address is never confirmed is deleted after seven days — or up to fourteen, if a confirmation link you asked for is still waiting to be used. If it began as a guest account with progress on it, the progress stays with the guest account and only the unconfirmed address is removed.

The sign-in and activity records described in section 2 are kept for as long as the account exists. They are how we tell that an account is still in use, so there is no shorter window after which they expire; they are deleted with the account, whether you delete it yourself or the sweep above reclaims a guest account.

Each data export (section 8) leaves a record of when it started, when it finished and how large the archive was. That record is what enforces the limit of one export every 24 hours. We keep each one for a year, after which a daily sweep deletes it, and it is deleted with the account if that comes first. The records are part of the export itself, so the archive lists the exports taken before it.

The notification delivery records described in section 2 are kept for 90 days, then a daily sweep deletes them. Our ntfy server keeps each message for 12 hours, so the app can fetch it, and then discards it. A channel you remove from your notification settings is deleted at once. A channel that stops working — you blocked the bot or sent it /stop, your browser withdrew the subscription, or a reminder email could not be delivered or was reported as spam — is kept, marked disconnected, for 30 days so the settings page can tell you what happened, and is then deleted.

Your answers — spoken, typed or dictated, with the silences your browser measured — and their evaluations are kept until you delete them — each answer on its own, once it has finished being evaluated — or until you delete your account. Recordings are never kept at all, and neither is the text of an answer the moderation check flags. Key points written for a question are kept with the question; they hold nothing of yours.

Chats are kept until you delete the conversation — a conversation is deleted whole, with every message in it — or your account. The recording of a spoken message and the audio of a reply read aloud are never kept, and the record of your agreement to the natural voice's notice is kept until you delete your account. If you cannot open a conversation on the site to delete it, because its question or lesson has since been retired from the catalog or the chat is not switched on for you, write to us and we will delete it.

Your credit ledger and promo-code records are kept for as long as your account exists. A promo credit expires 30 days after you redeem it; its expiry is recorded in the ledger rather than deleted from it.

The usage record of each AI call is kept for 400 days, then a daily sweep deletes it. If you delete your account sooner, the record loses its link to you and the rest — which holds no content — stays until then, so that our costs still add up.

While your account exists, your purchases are kept with it. When you delete it, each purchase is kept without the link to you: the account and your Telegram user id are removed from it, unless we still owe you a refund for a payment we could not credit, in which case the Telegram user id is kept until that refund is made. What remains is an accounting record: the reference, the pack, the amount, the dates, the ids Telegram gave the payment and the version of the terms you accepted. Georgian tax law requires such records to be kept for three years after the end of the tax year they belong to (Tax Code of Georgia, Article 43). We keep them until the end of the fourth calendar year after the purchase, or after its refund if that came later, which also covers the one extra year the same Code allows for a tax assessment, and then a daily sweep deletes them. An invoice you never pay is deleted after seven days.

Operational logs are kept for 7 days by Railway and 30 days on our own monitoring server, then deleted. The bot-protection counts described in section 2 never reach them: they live only in the server's memory and are gone at its next restart.

Deleted data can persist in our encrypted backups for up to 30 days after it leaves our live database: the database can be recovered to any point in roughly the last 27 days, and a nightly encrypted archive is kept for 30. After that it is gone. Backups are used only to recover the service from a disaster.

8. Your rights and choices

You can access and update your account details, manage your active sessions, enable or disable two-factor authentication, and reset your learning data — progress, schedules, quiz history and lesson progress — from your account settings.

Account deletion is available as a self-service action. Deleting your account (via your account settings, which calls DELETE /api/users/me) cascades: it removes your account record, your credentials, sessions and two-factor secrets, the sign-in and activity records described in section 2, the record of your data exports, your notification settings, connected channels and delivery records, and every piece of learning data keyed to the account — progress, schedules, review history, quizzes, the roadmaps you started, your progress through lessons and the lessons you added, and the answers you wrote yourself. It also removes your answers with their evaluations and the answer notice you agreed to, your chats and the natural voice's notice you agreed to, and your credit ledger and promo-code records; any credits left on the account, paid ones included, are forfeited, as the Terms of Service say.

Deleting asks you to prove the account is yours, and three proofs are accepted: your password, a code from your authenticator app or one of your recovery codes, or — if you signed in with a provider and have neither — a single-use link we email to the address on the account. Exactly one of the three, each time: being signed in is not on its own enough to delete an account.

Four things survive that deletion. The first is deliberate: bug reports and feedback you filed stay, so the problems they describe can still be fixed, but they are stripped of everything that links them to you: the account reference is cleared, so nothing left on the report points back to you, and we stop emailing about it. If you need the report itself removed rather than unlinked, ask us and we will delete it.

The second and third are records we must or need to keep: the record of each purchase you made, kept without the link to you for the tax-record period in section 7, and the usage records of the AI calls you made, kept without your account id until their 400 days are up.

The fourth is outside our database. If a reminder email to your address bounced permanently or was reported as spam, Amazon SES keeps the address on its suppression list, so that no mail is sent to it again. Deleting your account does not remove it from that list; ask us and we will.

Two things outlast the deletion briefly. Our internal records of the deletion itself, which carry your email address so every step can be completed, are removed about a week after the deletion completes, and if a step fails it is retried automatically until it does. Copies in our encrypted backups are gone 30 days after that, as section 7 describes.

A guest account cannot use self-service deletion. It has no email address and no password, so there is no way for it to prove the account is yours. To walk away from one, clear KataJob's cookies and site data from your browser: the account becomes unreachable, and it is deleted once it has gone unused for 180 days, as section 7 describes.

Depending on your location you may also have rights to access, correct, export, or restrict processing of your data. Export is self-service: your account settings will build you a ZIP archive of what the account holds — your profile and preferences, the days it was active, its security metadata (which sessions exist, when you last signed in, whether two-factor is on), all of your learning data including the answers you wrote out, your evaluated answers with their evaluations, your chats with the natural voice's notice you agreed to, your credit ledger, promo codes and purchases, your notification settings and connected channels, and the reports you filed. The archive holds JSON and CSV files and a README that says what each file contains. It proves the account is yours first, with your password, a code from your authenticator app or one of your recovery codes, or — if you signed in with a provider and have neither — a single-use link we email to the address on the account. That link confirms only the export: it cannot delete the account or change anything on it, and the link a deletion sends cannot be used to export. The README states the archive's omissions: the password hash, the two-factor secret and recovery codes, the value of any session or email token, the address and keys of a browser push subscription, the name of an ntfy topic, the bytes of files attached to a report, the usage records of AI calls (they hold no content, and the answers and chat messages they measured are in the archive unless they have been deleted), and server logs. One export is available every 24 hours; for anything else, write to [email protected].

The remaining rights have no button and are handled by request: correction, restriction, and a copy of the attachment bytes the export leaves out. Write to [email protected] and we will fulfil it.

9. Security

We protect your data with measures appropriate to its sensitivity: passwords are stored only as salted BCrypt hashes, two-factor secrets are encrypted at rest with AES-256-GCM and recovery codes are kept as hashes, the cookie that keeps you signed in is HttpOnly and same-site, repeated failed sign-ins trigger a temporary lockout, and requests are rate limited. No system is perfectly secure, but we aim to apply current good practice and to fix issues promptly.

10. Children

KataJob is intended for use by adults preparing for technical interviews and is not directed to children. You must be at least 16 to use it, and we do not knowingly collect data from anyone younger. Buying credits is for adults only: you must be 18 or older. If you believe a child has given us data, write to us and we will delete it.

11. International transfers

The providers named in section 6 all operate across multiple regions: Railway hosts the application and database, Cloudflare serves every request from whichever of its locations is nearest you, Resend and Google carry our mail, Amazon SES carries reminder email, offers email and purchase receipts, Telegram and your browser's push service carry the notifications you turn on, and ntfy.sh carries the signal that wakes the ntfy iOS app. OpenAI evaluates answers, replies to chat messages, and turns what you say into text and reads replies aloud when you talk in the natural voice, in the United States, and Telegram handles payments in Stars. Your data may therefore be processed outside the country you are in. Where required, we put appropriate safeguards in place for international transfers. Details are available on request.

12. Changes to this policy

We will update this policy as the product and our data practices evolve. Material changes will be communicated before they take effect, and the "last updated" date below will change.

13. Contact

Questions about this policy or your data can be sent to [email protected], which reaches the person who runs the site — section 1 says who that is. This policy is governed by the law of Georgia: the country, not the US state.