Privacy Policy

EdgeCue is designed from the ground up as a confidential, privacy-first, fully on-device AI cognitive rehearsal and speech teleprompter application for interview practice, study, and presentation rehearsal.

1. Architectural Privacy Guarantees

2. Information We Process & How It Is Handled

A. Microphone Audio (android.permission.RECORD_AUDIO)

Microphone audio is accessed solely during active, user-initiated practice sessions to perform real-time speech-to-text transcription and detect spoken interview questions.

• RAM-Only Ephemeral Lifecycle: Spoken audio is digitized into 16 kHz 16-bit mono PCM float arrays in volatile RAM, processed through a zero-allocation circular ring buffer (CircularFloatRingBuffer), fed to the on-device Silero VAD and Sherpa-ONNX Zipformer neural network, and immediately overwritten in memory upon utterance endpointing. Audio is never written to persistent flash storage, cache directories, or external storage.
• Foreground Service Notice: Continuous audio capture operates strictly under Android's FOREGROUND_SERVICE_MICROPHONE with an ongoing system notification featuring an accessible, single-tap "Stop Assistant" action button that terminates microphone capture immediately.
• Recording Consent & Lawful Use Notice: Before microphone access is initiated, the Application presents a prominent disclosure and pre-recording confirmation notice reminding users of their legal responsibility to comply with applicable local, state, provincial, and national wiretapping and recording consent laws (including US Federal 18 U.S.C. § 2511, all-party consent state statutes, Canadian Criminal Code s. 184, and Mexican Federal Penal Code Art. 211 Bis).

B. Local Session History & Transcripts

By default, session history is not retained. If and only if you explicitly enable the "Save Transcripts & Q&A History" toggle in Settings, practice transcripts and generated cue talking points are saved to a private, sandboxed SQLite database (interview_assistant.db) within internal app-specific storage (/data/data/com.edgecue.app/databases/).

C. Retrieval-Augmented Generation (RAG) Document Privacy (CV & Job Description)

When you provide a Curriculum Vitae (CV), resume, or job description to personalize your practice rehearsal, document text is parsed and indexed 100% locally on your device storage.

• Zero Cloud Ingestion: Document indexing and retrieval do not upload source files, chunks, or embeddings to AI services. A question or answer you explicitly report may include excerpts derived from those documents.
• Full User Control: You can view, replace, or permanently delete all indexed RAG documents at any time from the Settings screen.

D. Floating Teleprompter HUD (android.permission.SYSTEM_ALERT_WINDOW)

EdgeCue provides an optional floating teleprompter overlay (FloatingHudOverlay) that hovers over practice applications or coding documents during rehearsal.

• Low-Glare OLED Display: Designed with high-contrast, low-glare styling to minimize ocular fatigue during extended rehearsal loops.
• Mascot Pill: Features the Dachshund mascot icon and "EdgeCue" status label in a compact accessible pill.
• Content Safety Reporting: Includes an in-app "Report" action (ReportAiContentDialog) allowing users to submit inaccurate, biased, or unsafe responses through FormSubmit to the developer without opening another app.

E. On-Device AI Model Weights Provisioning (FOREGROUND_SERVICE_DATA_SYNC)

To enable offline, zero-latency inference, the Application allows you to download open-weight neural model files (Silero VAD, Sherpa-ONNX Zipformer, and Google LiteRT Gemma 4).

• Direct Encrypted CDN Downloads: Model assets (~2.6 GB to 3.1 GB) are retrieved directly over encrypted HTTPS from Hugging Face (huggingface.co / hf.co). EdgeCue does not route transfers through intermediate proxy servers and collects no connection metadata.
• Cryptographic Checksum Verification: All downloads are verified locally via streaming SHA-256 cryptographic checksums and exact byte length checks before being marked ready for inference.
• Full Offline / Airplane Mode: Once models are downloaded, speech recognition and answer generation work in Airplane Mode. Sending reports requires internet access.

F. User-Initiated Support & Feedback (Optional)

The Application includes an optional "Send Feedback" feature in Settings. Feedback is never transmitted automatically or in the background.

When initiated, the Application composes a plain-text/Markdown template and launches your device's native email client via Intent.ACTION_SEND addressed to hbenitez@hectisai.com. You have complete transparency and control to review, edit, or delete any content before deciding to send.

You may optionally choose to attach diagnostic metadata (app version, hardware SoC tier, model verification status) and/or recent session transcripts by explicitly checking in-app opt-in boxes. These options are disabled (unchecked) by default. Any feedback submitted is used strictly by the developer to resolve technical defects and is not sold or used for advertising. Your email provider and the developer's email provider process the message for delivery and storage.

G. In-App AI Content Reports (Optional)

When you tap Submit report, EdgeCue sends the displayed question, AI answer, selected reason, optional comment, and app version over HTTPS to FormSubmit, which relays the report to himmlerbenitez@gmail.com. FormSubmit and Google (Gmail) process the report for delivery and storage. Text may contain personal information you supplied or excerpts derived from your CV. The app does not attach raw audio, complete history, source documents, device diagnostics, or your contact details. Avoid adding sensitive information.

Reporting is optional and separate from session history. The app previews the text and shows whether the service accepted the request; acceptance does not guarantee inbox delivery. New reports are not saved to the local report database or queued for automatic retries. If a request fails, you can retry while the form remains open. Sending requires internet access.

Reports are used to investigate unsafe or inaccurate outputs and improve safety handling, not advertising. Developer-held emails are retained while an issue is investigated and removed when no longer needed, or on a valid deletion request, subject to legal retention obligations. Clearing history or uninstalling does not delete sent emails. Request deletion at hbenitez@hectisai.com or himmlerbenitez@gmail.com with an approximate submission date and enough information to locate the message. Older versions may have saved local reports; removing app storage removes those local records.

FormSubmit may receive connection metadata such as IP address. Its FAQ states that unconfirmed form submissions are retained for up to 30 days. Provider handling and retention are also governed by FormSubmit's privacy policy, FormSubmit's FAQ, and Google's privacy policy. Processing may occur outside your country, including for users in the United States, Canada, and Mexico. We do not promise a particular hosting region or that clearing app data removes provider-held copies.

3. Data Sovereignty & User Deletion Rights

You maintain absolute ownership and control over all local data generated by the Application:

• Clear Interview History: Navigate to the Activity tab (or Settings) and select "Clear History" to permanently and irreversibly purge all stored interview Q&A records and latency telemetry from the local database.
• Purge RAG Documents: Navigate to Settings -> Resume & Job Description (RAG) and tap "Clear Documents" to wipe all indexed CV and job description chunks.
• Purge AI Model Assets: Navigate to System & Diagnostics and select "Purge Downloaded AI Models" to instantly delete all local neural weights (~3.1 GB) from your device flash storage without uninstalling the app.
• Application Uninstallation: Uninstalling EdgeCue via Android OS settings immediately and irreversibly destroys all sandboxed databases, preferences, cache files, and downloaded model weights.

4. Hardware Acceleration & Benchmark Disclaimer

EdgeCue is optimized for Android devices running Android 13 to Android 17 (API 33–37):

• Google Tensor EdgeTPU / NPU (Pixel 8 / 9 / 10 Pro): Direct southbound native acceleration via bundled libLiteRtDispatch_GoogleTensor.so shims. Performance metrics published in documentation were recorded on a physical Google Pixel 10 Pro XL test unit; real-world performance will vary based on hardware model, thermal throttling, and operating environment.
• Qualcomm Snapdragon & Samsung Exynos: OpenCL and Vulkan mobile GPU acceleration with automatic dynamic sequence caching.
• No Guaranteed Outcomes: The Application is an assistive rehearsal tool. The developer does not guarantee specific token inference rates on all devices, nor does the developer guarantee employment, interview scores, or hiring outcomes.

5. Third-Party Trademarks & Nominative Fair Use

Google, Android, Google Pixel, Tensor, and Gemma are trademarks of Google LLC. Qualcomm and Snapdragon are trademarks of Qualcomm Incorporated. Samsung and Exynos are trademarks of Samsung Electronics Co., Ltd. All trademarks, logos, and brand names are the property of their respective owners. Reference to these trademarks is strictly nominative and descriptive to identify hardware compatibility and open-weight model dependencies. EdgeCue is an independent application and is not affiliated with, sponsored by, or endorsed by Google LLC, Alphabet Inc., Qualcomm, Samsung, or any of their affiliates.

6. Children's Privacy (COPPA & Google Play Families Policy)

EdgeCue is professional interview preparation software intended solely for adults aged 18 and older preparing for job, collegiate, or technical interviews. We do not knowingly target, solicit, or collect information from children under 13 (or under 16 where applicable by law).

7. Application Security & Defenses

• Global HTTPS Enforcement: The Application enforces a strict Android Network Security Configuration (res/xml/network_security_config.xml) that globally blocks all unencrypted cleartext HTTP traffic.
• Zero Dynamic Code Execution: EdgeCue loads only passive numerical neural weights (.onnx and .litertlm) into verified native runtimes. It never utilizes dynamic class loading (DexClassLoader) or downloads executable binaries.
• Production Log Stripping: All verbose, debug, and informational logs are stripped at compilation time via ProGuard/R8 to prevent transcript or audio telemetry leakage to Android system logcat.

8. International Compliance & Governing Law

This Privacy Policy is governed by the laws of the United Mexican States (Mexico). Depending on applicable law, you may have rights to access, correct, or delete personal information in optional reports and support messages. Contact the developer to make a request. Optional messages may be processed across borders by the receiving providers. EdgeCue does not create acoustic voiceprints or biometric templates.

9. This Website (edgecue.hectisai.com)

This website uses Vercel Web Analytics to count page views in aggregate. It does not set cookies or use cross-site tracking. According to Vercel, visitors are counted with a short-lived hash of request data that is discarded within 24 hours, and no personal profile is built. The website does not collect form submissions; the beta sign-up buttons send you to Google Groups and Google Play, which are governed by Google's own privacy policy.

10. Policy Updates & Inquiries

We may periodically update this Privacy Policy to reflect application improvements or legal requirements. Revisions will be published at this public URL with an updated "Last Updated" date.