How READSmart handles your data

READSmart is a research project led by Cyrill D. Tapales, MSIT of Eastern Visayas State University, Main Campus. It builds and tests an automated way to measure how fluently pupils in Grades 4 to 6 read aloud, in English and Filipino, using the Phil-IRI (DepEd Order No. 14, s. 2018).

The project lead is responsible for how READSmart handles personal data. Each partner school decides which of its teachers and pupils use READSmart, and its READSmart administrator manages their accounts. How to reach us is at the end of this page.

• Pupils who sign in with an account their teacher made, and their parents or guardians.
• Teachers and school administrators who manage classes, check readings and review stories.
• Visitors to the public demo, who need no account. The demo privacy notice has the full details.
• Research participants, whose readings are part of the study.

Teachers and school administrators

• Name, email address, school and an optional username.
• A password, stored only as a salted hash we cannot read back.
• A teacher's join request, the optional note sent with it, and who approved it and when.
• The classes each teacher is assigned to, and the scores, levels and messages they enter.

Pupils

• A pupil label and a username, which the teacher creates. By default the label is the first name and last initial (for example “Ana R.”) and the username is made from the first initial and surname. A teacher can change both. When a teacher pastes a class list, full names stay on the teacher's device: READSmart saves only the label and the username.
• Grade, class, school, and the reading-buddy picture the pupil chooses.
• A password. At first the teacher gives a temporary one on a printed login slip, which the pupil replaces when they first sign in.
• Pupils have no email address in READSmart and are never sent email.

Readings

• The audio of the pupil reading a passage aloud, which passage it was, the date and time, how long it lasted, and the recording format and quality the browser reported.
• For a reading check: when the pupil started and sent it, and whether a teacher allowed another try.
• Results worked out from the audio: a transcript, the time of each word, words correct per minute (WCPM), word-reading accuracy, a calibrated version of those two numbers, and practice reading-range estimates.
• The pupil's answers to the comprehension questions, whether they looked back at the story, and the score.
• Teacher scores, reading levels and messages to the pupil. We also keep the values READSmart filled in before the teacher checked them, so the study can measure how often teachers change them.

Practice stories that pupils ask for

• The topic the pupil typed or chose, the story language, and the progress of the request.
• The finished story, its title, comprehension questions and picture, plus the teacher's decisions about it (approved, hidden, shared or retired).

Records we keep to run the service

• Consent decisions: implied consent when a pupil account is made, and any refusal or withdrawal the school records.
• An audit log of important actions, such as accounts created, passwords reset, login slips printed, checks opened and consent recorded. It holds counts and record numbers, not recordings or text.
• Usage and cost records: how many readings were processed, audio length, processing times and outcomes. These hold no names, audio or text.
• Security records, such as sign-in attempt counts. We store only a scrambled form of the network address, never the address itself.
• When a teacher last opened the notification bell.

What we don't collect

We don't collect location, contacts, photos or video, and we don't use advertising or analytics trackers. Recordings are used to measure reading, not to recognize who is speaking.

• To let pupils read aloud, practice and see their own progress.
• To measure reading fluency and comprehension, and to help teachers review each reading, choose a Phil-IRI level, print Phil-IRI forms and see class statistics.
• To run reading checks at home, including the time limit.
• To make practice stories that pupils ask for, and to let teachers check them first.
• To keep accounts secure, prevent misuse, fix problems and keep costs within the project's budget.
• For the research study described below.

We don't sell personal data, share it for advertising, or use it for any purpose not listed here.

Speech processing

Each reading goes through open speech-recognition software that READSmart runs in its own container on GPU servers rented from RunPod: Silero VAD finds the speech, OpenAI Whisper (the downloadable model, size “medium”) writes the transcript, and WhisperX lines each word up with the audio. The audio is not sent to a speech-recognition company's service.

The container fetches the audio through a private link that works for at most 15 minutes, along with the passage text, grade and language. It never receives a name, username or email, but the storage path and record numbers it uses contain internal numbers for the school, reading session and pupil. It deletes its copy of the audio as soon as the reading has been processed.

What the scores mean

• READSmart compares the transcript with the passage and highlights words that differ. These highlights are text differences, not a judgement of pronunciation. READSmart does not detect mispronunciations.
• A small calibration formula corrects the automated WCPM and accuracy toward teachers' scores, using grade and language. It is fitted from results and teacher scores, never from voices.
• Comprehension answers are marked against an answer key. No AI marks them.
• READSmart only suggests a Phil-IRI level. The teacher listens and decides. No decision about a pupil is made by a computer alone.

Practice stories written by AI

When a pupil asks for a story, READSmart first checks the topic. It must be 2 to 60 letters and at most 8 words, and topics with names, personal details or unsafe words are refused before anything is saved. The story is then made through Cloudflare Workers AI:

• Mistral Small 3.1 writes the story from the topic, grade and language.
• Llama 3.3 70B gives it a title, checks its facts and writes comprehension questions.
• Llama Guard 3 checks the story for unsafe content.
• FLUX.2 [klein] 9B draws a picture from a short scene description, copying the drawing style of two of READSmart's own sample pictures, and Llama 4 Scout checks the picture against the story.

Only the writing model receives the pupil's topic, and no model receives the pupil's name or account. Cloudflare states that it does not use customer content to train AI models. Our AI Gateway keeps usage records (the step, grade, language and a random request number) without the prompts, stories or pictures. The picture step can't go through the gateway, so it goes to Workers AI directly, and READSmart counts those picture requests in its own records instead.

A teacher must approve every story before any pupil can read it. The teacher sees warnings from the fact check, can use a plain placeholder instead of the picture, and can hide or retire the story. A topic that passes the first check is saved with the request even if no story is made.

The sign-in page says that signing in means you agree to this policy. When a teacher makes a pupil account, READSmart records the pupil's consent as “implied by app use”. Practice recording and practice stories are then on by default, and the school can turn practice or stories off.

A parent or guardian can tell the school at any time that they don't want READSmart used for their child. The school's READSmart administrator records the decision; a teacher can pass the request on.

• A refusal stops new readings, reading checks, practice, practice stories and teacher messages for that pupil. Readings already made are kept.
• A withdrawal stops the same things and also deletes the readings made under that consent: the audio, transcripts, results, teacher scores and messages, and comprehension answers. The pupil's account, the consent record itself and content-free usage and audit records are kept. Practice stories the pupil asked for are also kept. Ask us if you want those removed too.

Consent records can't be edited or deleted, so there is always a record of each decision. For the research study, the school also collects signed consent forms from parents or guardians. The study relies on those forms, not on implied consent.

READSmart's main database is in Cloudflare's Singapore data center, and audio and pictures are stored in Cloudflare's Asia-Pacific region. Cloudflare serves pages from the data center nearest to you. RunPod's servers may be in other countries. Data may therefore be processed outside the Philippines. When the speech container starts, it downloads the model files from their public sources. That download includes no pupil data.

Readings that come in while a school uses READSmart are part of the study. When the study ends, the project lead will delete or de-identify the data it no longer needs, and will tell partner schools what it keeps.

• Passwords are stored only as salted PBKDF2-SHA256 hashes (600,000 rounds). Temporary passwords are sealed with AES-256-GCM.
• Sign-in uses encrypted, HttpOnly cookies, checked against the account on every request. Sign-in attempts are rate-limited, and changing a password signs out older sessions.
• Recordings go straight from the browser to private storage through upload links that expire after 15 minutes. They can be played back only through READSmart, after checking that the person may hear them.
• Every request is limited to the person's role, school and classes. Staff accounts need an administrator's approval.
• Pages cannot be framed by other sites, and only READSmart's own pages may use the microphone.
• The speech container reaches READSmart only with a secret token and never connects to the database directly.

READSmart uses only the cookies it needs to work. There are no advertising or analytics cookies.

READSmart also stores a few things on your device:

• Session storage (cleared when the tab closes) remembers the chosen microphone, whether a practice reading has started, and the demo session's details.
• A recording is saved on the device before it is sent, so it isn't lost if the connection drops. It is deleted once READSmart receives it or the attempt expires.
• The app's offline helper saves public pages and app files so READSmart can load without a connection. It never saves signed-in pages or personal data.

Under the Philippine Data Privacy Act of 2012 (Republic Act No. 10173), you have the right to:

• be told how your personal data is used, which is what this policy is for;
• see the personal data we hold about you or your child;
• have wrong or incomplete data corrected;
• object to its use, or have it blocked or deleted;
• receive your data in a common electronic format;
• be compensated for damage caused by the wrong use of your data;
• complain to the National Privacy Commission.

Pupils and parents can start with the pupil's teacher or the school's READSmart administrator. The teacher can correct a pupil's label or username, and the administrator can record a refusal or withdrawal. For anything else, including copies of data or deleting data a withdrawal doesn't cover, contact the project lead. We may need to confirm who you are, and that you are the pupil's parent or guardian, before we act on a request.

• Pupils sign in with an account from their teacher. READSmart is not open for children to sign up by themselves.
• Teachers can open a reading check that pupils do at home. Before starting, the pupil tests the microphone: that test recording stays on the device and is never sent. The passage stays hidden until recording starts, and a timer shows the time left.
• No one watches a reading at home. The timer is only a guide, so parents are welcome to help the pupil get set up in a quiet place.
• When pupils ask for practice stories, they should not type names or other personal details into the topic. READSmart refuses topics that look like they contain them.
• The browser asks permission before READSmart can use the microphone, and READSmart records only while the pupil is reading.

READSmart is part of a study that checks how well automated scores agree with teachers' Phil-IRI scores in English and Filipino. The study compares each recording's automated results with the scores teachers give it.

• The research sample is covered by signed consent forms from parents or guardians, collected through the school.
• The research team analyses copies with names and pupil labels removed. Published findings report only group results and code teachers as T01, T02 and so on. They never include audio, names, labels or school identifiers.
• The research team used an AI assistant (Anthropic's Claude) to help read scanned Phil-IRI scoring sheets for the study, then checked the results.
• Readings done at home can be told apart from readings done with a teacher, because reading at home differs from the Phil-IRI's teacher-led method.
• Demo recordings are never part of the research.

The demo lets anyone try READSmart without an account. It asks for no name, email, school or class. The server deletes the audio as soon as it has been processed, and the result is shown only after deletion is confirmed. The transcript and result expire three hours after the session starts. Demo data is never used for classroom records, research, model training or public samples.

Read the full demo privacy notice

We update this page when READSmart changes how it handles data, and change the date at the top. If a change affects how pupils' recordings are used, we will tell partner schools before it takes effect.

Questions, requests and complaints about privacy go to the project lead:

Cyrill D. Tapales, MSIT
Eastern Visayas State University, Main Campus
cyrill.tapales@evsu.edu.ph
+63 950 752 8822

If you are not satisfied with our answer, you can contact the National Privacy Commission.