1. Information We Collect
We collect information in three ways: information you directly provide, information collected automatically when you use the Service, and information received from third-party integrations.
A. Account & Profile Data
When you register, we collect your name, email address, password hash, target language selections, native language, current CEFR skill level, learning motivation, preferred practice schedule, and timezone.
B. Voice & Audio Data
When you initiate a WebRTC voice call, use microphone input, or perform pronunciation practice, we collect and process raw audio streams, audio recordings, speech waveforms, phonetic timing, acoustic features, and AI-generated transcripts of your spoken utterances.
C. Conversational & Learning Interaction Data
We collect all text chats, prompt queries, grammar corrections, quiz answers, spaced repetition system (SRS) flashcard reviews, mastery scores, streak statistics, journal entries, and guided conversation tree progression logs.
D. Technical, Device & Usage Telemetry
We automatically collect your IP address, browser user-agent, operating system, device hardware identifiers (including Pebble smartwatch pairing data and mobile push notification tokens), network connectivity metrics, WebRTC latency telemetry (StageTimer TTFB), and session cookies (__Secure-abjad.session in production and abjad.session in development).
E. Billing & Payment Data
Payments are processed directly by our payment processor, Stripe, Inc. Abjad does not receive, process, or store your full credit card numbers, CVVs, or cardholder banking credentials. We retain only Stripe customer identifiers, subscription tiers, renewal dates, and transaction receipts.
2. How We Use Data & AI Model Training
We process personal information for the following operational, educational, and developmental purposes:
- Providing the Tutoring Service: Operating interactive voice and text sessions, synthesizing audio replies via neural text-to-speech (TTS), evaluating pronunciation, generating grammar corrections, maintaining spaced repetition review queues, and personalizing difficulty.
- Account Management & Communications: Authenticating sessions, sending email verification, delivering weekly progress digests via Resend, and providing customer support.
- Billing & Access Control: Managing Abjad Premium subscriptions and enforcing feature gates.
- Platform Security & Optimization: Monitoring system stability, mitigating latency, preventing abuse or prompt injection attacks, and enforcing our Acceptable Use Policy.
Mandatory Artificial Intelligence & Voice-to-Voice LLM Training Notice
A primary and material purpose of collecting user interactions and audio on Abjad is to advance and train artificial intelligence models. Specifically, all User Content, voice call audio, audio streams, transcripts, and interaction data are utilized by Abjad Inc. to:
- Train, fine-tune, and align our proprietary autonomous Voice-to-Voice Large Language Model (LLM) agents, enabling end-to-end conversational speech comprehension and low-latency acoustic generation;
- Train and optimize Automatic Speech Recognition (ASR) systems for accented and multilingual speech;
- Train and enhance neural Text-to-Speech (TTS) models and voice synthesis engines;
- Train multimodal language models, semantic vector embeddings, and pedagogical scoring algorithms;
- Develop and commercialize next-generation voice AI products, developer APIs, and educational tools.
No Opt-Out: As stated in Section 5 of our Terms of Service, granting permission for AI model training is a non-negotiable requirement of using Abjad. If you do not wish to have your audio and data used for AI model training, you must not use the Service.
Pipeline Controls & Feature Settings: While you grant Abjad the overarching license and legal authority to use interactions and audio for AI training, our operational ingestion pipelines apply automated quality filters, PII sanitization, and safety deduplication. Specific developer or enterprise configurations (such as Bring-Your-Own-Key [BYOK] distillation settings) may apply additional technical routing controls prior to model ingestion without waiving the platform license.
3. Voice & Biometric Data Disclosures
Certain jurisdictions (such as Illinois under the Biometric Information Privacy Act, 740 ILCS 14/1 et seq., and California under the CCPA/CPRA) regulate biometric identifiers and voice recordings.
Educational & Acoustic Processing Notice: Abjad processes voice recordings and acoustic speech signals solely to conduct language tutoring, evaluate pronunciation, transcribe speech, and train machine learning models.
No Biometric Identification: We do not extract unique mathematical voiceprints, vocal biometric templates, or geometric voice signatures for the purpose of individual biometric identification, physical authentication, identity verification, or behavioral surveillance.
Retention Schedule: Raw audio files containing identifiable user speech are retained in secure cloud storage for up to three (3) years from the date of collection or until account deletion, whichever occurs first. De-identified acoustic training datasets, latent embeddings, and derived neural network model weights are retained indefinitely as part of our permanent model assets.
Storage & Destruction: Audio data is stored using industry-standard AES-256 encryption. When raw identifiable audio reaches the end of its retention schedule, it is purged from our production and backup storage systems using secure deletion protocols.
4. Legal Bases for Processing (GDPR / UK GDPR)
If you are located in the European Economic Area (EEA), the United Kingdom, or Switzerland, we process your personal data pursuant to the following lawful bases under the General Data Protection Regulation (GDPR):
- Performance of a Contract (Art. 6(1)(b) GDPR): Processing is necessary to deliver the conversational AI tutoring service, execute live speech and text interactions, maintain your learning state and spaced repetition history, and fulfill the core educational features agreed upon in our Terms of Service.
- Legitimate Interests (Art. 6(1)(f) GDPR): Processing is necessary for our legitimate interests in improving educational model accuracy, fine-tuning and benchmarking speech recognition and AI algorithms using de-identified interaction data, preventing abuse or prompt injection attacks, securing our infrastructure, and optimizing voice latency. Where required, we balance these interests against your fundamental privacy rights.
- Statutory Recording Consent (Local & US State Laws): For jurisdictions that require express all-party consent to record or transcribe electronic audio communications (such as US state wiretap laws), you provide affirmative consent by initiating live voice sessions and agreeing to our Terms of Service.
- Compliance with Legal Obligations (Art. 6(1)(c) GDPR): Maintaining financial records, tax compliance, and responding to lawful statutory requests.
5. Third-Party Sub-Processors & Sharing
We do not sell, rent, or trade your personal information to third-party data brokers or advertising networks. We share data only with trusted third-party sub-processors who perform essential services on our behalf under strict contractual data protection agreements:
| Sub-Processor | Category / Purpose | Location |
|---|---|---|
| Stripe, Inc. | Payment processing & subscription billing | United States |
| Supabase / PostgreSQL | Managed database, user profiles & session data | United States |
| Vercel Inc. | Web application hosting & edge network | United States |
| Resend, Inc. | Transactional emails & weekly progress digests | United States |
| OpenAI, LLC / DeepSeek | Upstream LLM completion & linguistic evaluation | United States / Global |
| ElevenLabs, Inc. / Microsoft Azure | Neural text-to-speech audio synthesis | United States |
6. International Data Transfers
Abjad Inc. is headquartered in the United States, and our primary servers and databases are located in the United States. If you access the Service from outside the United States, your information will be transferred to, stored, and processed in the United States.
Where required by applicable data protection law (such as the GDPR), we implement appropriate safeguards, including the European Commission’s Standard Contractual Clauses (SCCs), to govern cross-border data transfers.
7. Data Retention & The Irreversibility of AI Model Weights
We retain personal data only as long as necessary to fulfill the purposes outlined in this Privacy Policy, comply with statutory obligations, resolve disputes, and enforce our agreements.
Account Deletion: You may request the permanent deletion of your account and personal data at any time via your account settings or by emailing privacy@talkabjad.ai. Upon receiving a verified request, we will permanently delete or anonymize your name, email, credentials, raw identifiable audio recordings, and user-specific transcripts from our active production systems within thirty (30) days.
Critical Technical Disclosure: Irreversibility of Model Weights
When artificial intelligence and deep neural network models (such as our Voice-to-Voice LLM agents) are trained or fine-tuned, data inputs are converted into complex, non-reversible mathematical matrices, parameter weights, and latent representations.
These mathematical weights do not contain copyable records or retrievable documents of your personal data and cannot be “un-learned” or isolated after model training. Therefore, while Abjad will delete all identifiable raw data from our databases upon request, account deletion does not and cannot result in the deletion, roll-back, alteration, or destruction of any trained machine learning models, neural network weights, or derived algorithmic assets.
8. Your Privacy Rights (NJDPA / CCPA / GDPR)
Depending on your jurisdiction (including the State of New Jersey under the New Jersey Data Privacy Act, California under the CCPA/CPRA, and the European Union under GDPR), you have specific statutory rights:
- Right to Know & Access: You may request details about the categories and specific pieces of personal information we have collected about you.
- Right to Correction: You may request that we rectify inaccurate or incomplete personal data.
- Right to Deletion: You may request that we delete your personal information, subject to the technical limitations regarding trained model weights described in Section 7.
- Right to Data Portability: You may request an export of your learning data and account profile in a structured, machine-readable format.
- Right to Non-Discrimination: We will never discriminate, charge different rates, or deny services to you for exercising any statutory privacy rights.
To exercise any of these rights, please contact our Data Protection Officer at privacy@talkabjad.ai. We will verify your identity before processing your request and respond within the statutory timeframe (typically 30 to 45 days).
9. Children’s Privacy (COPPA Compliance)
The Service is designed and intended exclusively for individuals aged 13 and older. We do not knowingly solicit, collect, record, or maintain personal information or voice recordings from children under the age of 13 pursuant to the Children’s Online Privacy Protection Act (COPPA, 15 U.S.C. § 6501 et seq.).
If we discover that an account has been created by or on behalf of a child under 13, or that voice audio from a child under 13 has been transmitted through the Service, we will immediately terminate the account and permanently delete all associated records from our systems. If you believe a child under 13 has registered, please notify us immediately at privacy@talkabjad.ai.
10. Information Security Safeguards
We maintain rigorous administrative, technical, and physical security measures designed to safeguard your data against unauthorized access, loss, alteration, or disclosure:
- Encryption in Transit: All data transmissions—including WebSockets, WebRTC signaling, and HTTPS traffic—are encrypted using TLS 1.3 / DTLS 1.2.
- Encryption at Rest: Database records, transcripts, and audio storage buckets are encrypted using AES-256.
- Session Security: We utilize secure, HTTP-only, SameSite cookies with mandatory HTTPS in production environments.
- Access Controls: Access to user audio and interaction databases is restricted to authorized personnel on a strict, audited, least-privilege basis.
11. Policy Changes & Contact Us
We may update this Privacy Policy from time to time to reflect changes in our legal obligations, features, or artificial intelligence training methodologies. We will notify you of material modifications by sending an email notification or displaying an in-app banner prior to the changes taking effect.
Contact Privacy & Legal Operations:
Abjad Inc.
Attn: Data Privacy Officer
State of New Jersey, United States
Email: privacy@talkabjad.ai