Trellium Demo Privacy Policy
Document ID: TRELLIUM-DEMO-PP-2.1.0 Document version: 2.1.0 Effective date: October 26, 2026 Supersedes: version 2.0.0 (effective September 8, 2026)
The document version above is the identifier Sparked records when it registers your acceptance of this Policy. If you need to know which version applied to a particular Demo session, reference that identifier when you contact us.
This Privacy Policy explains how Sparked Innovations.AI (“Sparked Innovations,” “Sparked,” “we,” “us,” or “our”) collects, uses, discloses, retains, and otherwise processes personal information in connection with the invitation-only Trellium demonstration application, including portions of the application that may display the SparkEd name (collectively, the “Demo”).
Sparked Innovations.AI is the organization responsible for the personal information described in this Policy. Sparked is registered in Delaware and is located at 8 The Green Ste B, Dover, Delaware 19901, United States. You may contact us at kat@sparkedinnovations.ai or +1 949-826-1060.
Important voice and AI disclosure
Every voice-practice session in the Demo is recorded or otherwise captured, transcribed, and analyzed for vocal-expression signals. Scored sessions may be evaluated by a separate AI system to generate a formal coaching report; ungraded practice does not invoke that separate post-session evaluator or produce a formal learner report. Sparked’s authorized personnel and contractors may review session information. Other active Users in your Demo instance can view your name and eligible conversation transcripts and, after submitting their own human assessment, your saved AI overall and capability scores. Human assessment drafts, explanations, notes, and submitted ratings remain private to their reviewer. Sparked and its service providers may use identifiable conversations, recordings, transcripts, metadata, evaluations, and feedback to research, develop, train, test, validate, and improve SparkEd, Trellium, related products, and the artificial-intelligence and machine-learning systems used in or offered through Sparked’s products and services.
Audio may be used to train and evaluate speech, expression, and conversation models. Those models are not used to identify or authenticate any person. Sparked does not create a voiceprint, does not use voice data as a biometric identifier, and does not compare your voice against an identity database.
Sparked may retain and use de-identified information, aggregate findings, derived analytics, evaluation methods, and model artifacts that cannot identify you indefinitely. This product and model-improvement use is a required condition of Demo participation.
By accessing or using the Demo, you accept the Trellium Demo Terms of Service and this Policy, and you consent to the recording, transcription, expression analysis, automated evaluation, human review, and research and model-development uses described in them. That consent applies to each Demo session you start while it remains in effect.
This acceptance is intended to document your express consent where applicable law requires every party to a communication to consent to its recording, including when you participate from California. It does not provide consent for anyone else whose voice or information may be captured.
You may withdraw consent for future sessions by stopping use and contacting us. Because recording and analysis are essential to the Demo, withdrawal prevents additional voice-session participation. Withdrawal does not affect processing that occurred before it became effective or require deletion of information we may lawfully retain. You may separately request deletion as described in Section 11.
1. Scope; separate marketing-site policy
This Policy applies only to personal information processed through or for the Demo. It does not apply to Sparked’s public marketing website, lead-generation forms, public events, or unrelated services. The public website’s separate privacy notice governs those activities.
The public website privacy notice is not incorporated into this Policy and does not describe the Demo’s voice recordings, transcripts, expression analysis, scoring, raw archives, or product- and model-improvement uses. If the two notices appear to conflict about Demo information, this Policy controls for the Demo.
This Policy applies only to users in the United States. The Demo is not offered to people outside the United States or to anyone under 18.
2. Notice at collection
The following table summarizes the personal information Sparked collects when you use the Demo and why Sparked uses it. Retention is described in Section 10. More detail appears in the sections that follow.
| Information | Examples | Primary purposes | Retention |
|---|---|---|---|
| Account and identity information | Name, email address, user and account identifiers, password hash, account status, invitation information | Provision and secure the account; authenticate you; communicate about the Demo; administer the testing program | Up to 36 months |
| Consent and policy records | Accepted Terms and Policy document identifiers and versions, consent language, timestamps, IP address, user agent | Document and administer your agreement, recording consent, and policy changes; establish or defend legal claims | Up to 36 months |
| Voice, transcript, and session content | Your microphone audio, AI assistant audio, speech transcripts, messages, configured tool content, selected application WebSocket frames, timestamps, interruptions, mute and playback events | Provide the voice simulation; preserve and recover sessions; generate reports; quality assurance; support; safety; product and model improvement | Up to 36 months |
| Expression information and inferences | Hume expression measurements; probabilistic vocal-expression signals; annotations; communication-pattern inferences | Operate the simulation; provide coaching; score sessions; assess reliability; research and improve features, rubrics, and models | Up to 36 months |
| Evaluation and report information | Deterministic metrics, evidence citations, scores, confidence, rubric results, strengths, improvements, corrective scripts, practice plans | Generate and display coaching; assess Demo quality; support testing; research and improve products and models | Up to 36 months |
| Technical, network, and usage information | IP address, user agent, browser or device class, operating environment, session cookie, routes, request identifiers, configuration and model identifiers, chat identifiers, event sequences, timestamps, network class, latency, errors, and diagnostic data | Authenticate users; operate and secure the Demo; rate limit; diagnose errors; recover sessions; measure reliability and performance; improve the Demo | Up to 36 months |
| Support and feedback information | Support messages, bug reports, requested troubleshooting data, suggestions, and survey responses, including but not limited to ratings of voice audio clarity, voice naturalness, conversation flow, and response time, and free-text descriptions of anything that disrupted the experience and any further comments | Respond to requests; troubleshoot; protect the Demo; conduct research and product and model improvement | Up to 36 months |
| Security and audit information | Sign-in attempts, rate-limit records, account disablement, deletion activity, security events, and audit records | Prevent abuse and fraud; investigate incidents; demonstrate compliance; enforce agreements | Up to 36 months |
De-identified information, aggregate findings, derived analytics, evaluation methods, and model artifacts that cannot identify you may be retained indefinitely.
Sparked does not sell these categories of personal information or share them for cross-context behavioral advertising. Sparked does not use them to make legal or similarly significant decisions about you. Sparked discloses relevant categories to service providers and contractors for the purposes described in this Policy.
If Sparked wants to collect an additional category or use personal information for a materially different, incompatible purpose, Sparked will provide a new notice and obtain additional consent when required before doing so.
3. Personal information we collect
3.1 Information provided by you or for your account
Sparked collects information supplied when an account is provisioned, such as your name and email address. Sparked stores a password hash rather than your readable password.
Sparked also collects what you say during a voice session, written information you submit, support communications, and feedback. Your speech or submissions may incidentally reveal personal or sensitive information even when Sparked did not request it.
You must not speak or submit the name of any real patient, or any other information that could identify a real patient, at any time during your use of the Demo. This prohibition applies to full names, initials, dates of birth, record numbers, contact details, and any combination of details that could reasonably identify a specific person.
More broadly, do not provide protected health information, medical records, adverse-event or product-complaint information, confidential customer information, trade secrets, or another person’s personal information. The Demo is not configured as a HIPAA-compliant service, and Sparked does not offer a business associate agreement for it. The Demo does not scan, detect, or filter prohibited content. If prohibited information is submitted, Sparked and its providers will process it as part of the session until it can be identified and removed, unless retention is required by law.
3.2 Voice and conversation information
When you start a session, the Demo accesses your microphone and streams audio to Hume AI. The Demo or its providers may process and retain:
- microphone audio and AI-generated assistant audio;
- final and interim transcripts;
- user, assistant, system, tool, and lifecycle events;
- Hume chat and chat-group identifiers;
- scenario, configuration, version, voice, and model identifiers;
- timing, turn-taking, interruptions, pauses, mute state, and playback events;
- expression measurements and other numeric signal scores;
- selected application WebSocket frames; and
- event hashes, sequence numbers, completeness indicators, and evidence needed to verify the integrity of a completed session.
Hume expression measurements and Sparked’s derived inferences are probabilistic coaching cues. They are not verified facts about your emotions, intent, character, health, competence, or work performance.
Audio may be used to train and evaluate speech, expression, and conversation models. Those models are built to understand speech, vocal expression, and conversational dynamics in general. They are not used to identify or authenticate any person. Sparked does not create a voiceprint, use voice data to uniquely identify or authenticate you, or compare your voice against an identity database. Voice recordings are nevertheless personal and potentially sensitive, and Sparked treats them accordingly.
3.3 Scores, reports, and derived information
After a completed scored session, Sparked obtains authoritative Hume chat history and sends selected evidence to OpenAI Decisions for scoring, then to OpenAI Responses through TanStack AI for score-aligned coaching feedback. Ungraded sessions remain in personal history and can include a post-call survey, but are not sent to Sparked’s separate OpenAI evaluator for a formal score or report. A selected live voice model may still use Anthropic within Hume in either mode. The evidence may include transcript text, role and event identifiers, timestamps, up to six top vocal-expression signals for an event, deterministic session metrics, the scenario context, and the scoring rubric.
For scored sessions, Sparked’s systems may validate and derive information such as learner and assistant turns, questions, word share, filler words, speaking rate, turn duration, evidence citations, scores, confidence, coaching tier, strengths, improvement areas, corrective scripts, and practice recommendations. Sparked stores any resulting report and associated integrity and model-version information with your session history.
3.4 Technical, device, and usage information
Sparked and its providers automatically receive technical data when you access the Demo. This may include your IP address; browser, device, and user-agent characteristics; cookie and session identifiers; requested URLs; timestamps; referral and response data; network and browser class; latency; configuration and provider identifiers; usage events; diagnostic payloads; crash information; and security signals.
Sparked uses your IP address to authenticate sessions, apply rate limits, and protect the Demo against abuse. Sparked does not use it to determine or infer your location, and the Demo does not request precise geolocation from your browser.
Sparked uses an HTTP-only session cookie for authentication. The cookie is SameSite=Lax, applies across the application, and is Secure in production. Sparked does not use the Demo’s cookies or similar technologies for targeted advertising.
Operational logs and error reports may include identifiers, configuration details, expression signals, transcript or message content, settings, and tool content when necessary to diagnose a provider or application event. Sparked’s controls are designed to redact credentials, authentication tokens, API keys, passwords, and raw audio bytes before operational telemetry is transmitted, but no automated redaction is infallible.
3.5 Information from service providers
Sparked receives chat histories, transcripts, expression measurements, identifiers, audio status, and provider metadata from Hume; report output and API metadata from OpenAI; and infrastructure, error, security, and performance information from hosting and observability providers.
Sparked also collects the survey responses you choose to provide after a session, including but not limited to ratings of voice audio clarity, voice naturalness, conversation flow, and response time, and free-text descriptions of anything that disrupted the experience and any further comments. Free-text survey fields accept whatever you write and are not scanned or filtered.
4. Browser-local storage and recovery
The Demo uses browser storage to make voice sessions diagnosable and to recover completed work after ordinary connectivity or page failures. This information remains on the device and browser profile where it was created unless and until it is uploaded, removed by the Demo, cleared by you, or evicted by the browser.
4.1 Voice Event Archive
The browser stores a diagnostic archive of deserialized Hume SDK responses and lifecycle events in IndexedDB, including for failed or aborted attempts. Depending on the event, this archive may include transcripts, expression signals, settings, identifiers, and tool content. Encoded audio is excluded from this diagnostic archive.
4.2 Completion and evidence recovery queues
Separate IndexedDB queues may store completed session payloads, normalized event evidence, delivery state, and retry information until the server acknowledges receipt.
4.3 Raw-archive staging
For the Demo, private raw archiving is enabled. During an authorized session, the browser captures selected microphone audio, assistant audio, and application frames in browser-local IndexedDB. When the Demo uploads an authorized archive to Sparked-controlled storage, it uses encrypted network transport. Raw bytes do not leave the browser for that storage until normalized evidence is durably committed and the server authorizes post-session upload.
The browser packages authorized bytes into hashed ZIP archives and uploads them to private object storage. If capability discovery or upload fails, pending bytes may remain in browser storage for retry rather than being uploaded or discarded on an uncertain result.
To remove browser-local information, use the Demo’s Lock action after allowing pending recovery work to finish. You may also clear site data through your browser, but doing so may permanently discard unsynchronized sessions or evidence.
5. Server and provider storage
5.1 Sparked-controlled records
Sparked stores account, session, normalized evidence, transcript, expression, report, evaluation-integrity, deletion, and audit records in a U.S.-hosted PostgreSQL database. Completed sessions are associated with the authenticated user who owns them.
Sparked stores verified raw archives in private S3-compatible object storage. These archives may include microphone audio, assistant audio, and selected application frames. They support evidence integrity, diagnosis, recovery, authorized investigation, quality assurance, research, and product and model improvement. They are not used as the authoritative source for automated scoring and are not shown in the learner report.
5.2 Hume-controlled records
Hume processes live audio to transcribe speech, measure expression, generate the AI persona’s response, synthesize assistant audio, and retain chat history used for Sparked’s post-session scoring when scoring is selected. Hume’s chat history can include transcripts, expression measurements, system and tool events, configuration details, and audio sufficient for reconstruction. Sparked cannot use zero-data-retention for these Demo sessions because scored practice depends on retained Hume chat history and both practice modes use it for recovery and support.
Hume’s Chat History documentation explains that retained chats can include transcripts, expression measurements, and audio reconstruction. If an authorized Sparked operator requests a Hume audio reconstruction for a retained chat, Hume’s Audio Reconstruction documentation states that the reconstructed file may be stored indefinitely. Sparked does not automatically import a Hume reconstruction into its private archive and will limit any local operator copy to the approved support, investigation, or research task.
Hume’s published descriptions of model-improvement use are not fully uniform. Hume’s API Data Usage Policy states that Hume does not use customer-submitted API data to train its models or improve its service, while Hume’s EVI changelog has described a separate account-level “do not use for training” option for anonymized EVI data. Sparked has not enabled that opt-out for the Demo. Accordingly, Hume may use anonymized session data for its own model improvements to the extent permitted by Sparked’s account settings, Sparked’s agreement with Hume, and Hume’s then-current terms and policies. Contact Sparked if you want current information about the applicable setting.
5.3 OpenAI and Anthropic processing
OpenAI processes transcript evidence, scenario and rubric context, expression summaries, and deterministic metrics through Decisions to produce scores. Responses receives the scored evidence and decisions to generate coaching feedback. Sparked requests that Responses output not be stored as application state. Provider abuse-monitoring logs and other applicable retention can still apply. OpenAI's API data is not used for training by default; retention depends on the endpoint, account controls, and applicable exceptions. See OpenAI's data controls.
Anthropic may process live session content as a language-model provider selected within Hume in either practice mode. Its processing is governed by Hume's provider arrangements and the applicable provider terms and account settings.
5.4 Other provider-controlled records
Depending on the model variant and configured tools, Hume may route relevant session information to Google Gemini and to third-party search or tool providers. Those providers process the information under Hume’s provider arrangements and applicable terms. Hosting, database, storage, security, and observability providers may also maintain infrastructure records needed to perform their services.
Because Sparked necessarily relies on external vendors, those vendors process information under their own published policies, retention practices, and account settings, which Sparked does not control and which may change. Provider-controlled copies may persist independently of Sparked’s own records when permitted or required under provider contracts, account settings, abuse-prevention rules, security requirements, legal obligations, backup cycles, or technical limitations. Sparked will direct service providers to delete or return personal information when required by applicable law and its contracts, but Sparked cannot guarantee that a provider can remove information already de-identified or incorporated into a non-identifying model artifact.
6. How we use personal information
Sparked uses the information described above to:
- Provide the Demo. Provision accounts; authenticate users; connect voice sessions; transcribe speech; generate AI responses; preserve session history; and, for scored sessions, provide scores, coaching, and reports.
- Record and analyze sessions. Capture audio, events, and metadata; measure probabilistic vocal-expression signals; calculate communication metrics; and evaluate scored sessions against versioned rubrics.
- Recover and verify evidence. Resume interrupted delivery; reconcile browser and server records; verify archive integrity; diagnose incomplete sessions; and prevent silent alteration or replay of evidence.
- Support testers. Respond to questions; investigate bugs; reproduce failures; and provide technical or program support.
- Secure and administer the Demo. Protect accounts; enforce invitations; rate limit sign-ins and endpoints; detect and investigate misuse; audit deletion; enforce the Terms; and maintain service reliability.
- Research, develop, train, and improve. Conduct authorized human and automated review; create annotations and evaluation datasets; compare scenarios, models, prompts, rubrics, and features; train, fine-tune, test, validate, and improve the artificial-intelligence and machine-learning systems used in or offered through Sparked’s products and services, including speech, expression, and conversation models trained and evaluated on session audio; and develop new or improved SparkEd and Trellium products.
- De-identify and aggregate. Remove or transform identifiers, combine results, measure program and product trends, and create non-identifying research, reliability, and performance insights.
- Communicate. Send account, security, testing-program, support, and policy notices. The Demo does not use personal information for cross-context behavioral advertising.
- Comply with law and protect rights. Respond to lawful process; preserve records required by law; establish, exercise, or defend claims; protect the rights and safety of testers and others; and support a corporate transaction.
7. AI, product, and model improvement
Product and model improvement is a core purpose of the Demo, not an incidental use. Sparked and its service providers may use your recordings, transcripts, expression signals, interaction metadata, annotations, reports, and feedback for authorized human review and to create or evaluate datasets, rubrics, prompts, scoring systems, product features, and artificial-intelligence and machine-learning models, including models incorporated into Sparked’s products and services.
Sparked does not limit this use to systems operated only inside Sparked. The Demo depends on external vendors — including voice, language-model, hosting, storage, and observability providers — and those vendors process information under their own terms, policies, and account settings as described in Section 5.
Audio may be used to train and evaluate speech, expression, and conversation models. Those models are not used to identify or authenticate any person, and Sparked does not create a voiceprint or use voice data as a biometric identifier.
Sparked may compare provider and model performance, test failure modes, validate evidence and citations, investigate bias or reliability, and use selected examples to train or evaluate systems. Access is limited to personnel, contractors, and providers with a business need for these activities.
Where practicable for a research or training purpose, Sparked will remove direct account identifiers or use aggregate information. De-identification reduces but does not eliminate all privacy risk. Sparked will maintain measures designed to prevent information treated as de-identified from being associated with you and will not attempt to re-identify it except to test or improve de-identification safeguards or as permitted by law. When Sparked discloses information that it maintains as de-identified, Sparked will take reasonable steps to require the recipient to maintain it in de-identified form and not attempt to re-identify it, except as permitted by law.
Sparked may publish de-identified excerpts, clips, statistics, aggregate findings, and case studies that do not identify you. Sparked will not publicly identify you with your session content or use your identity in advertising or endorsements without your separate written permission.
A later deletion request cannot necessarily remove the influence of an example from aggregate statistics, evaluation methods, or already-trained model parameters when those artifacts do not contain information capable of identifying you. Sparked may continue to retain and use those non-identifying artifacts indefinitely.
8. How we disclose personal information
Sparked may disclose personal information to the following recipients for the purposes described in this Policy:
| Recipient | Information and role |
|---|---|
| Hume AI and its subprocessors | Live and retained audio, transcripts, expression measurements, chat and configuration data, session events, and information sent to configured language models or tools for voice simulation and chat-history services |
| OpenAI | Transcript evidence, expression summaries, metrics, scenario context, rubrics, decisions, and API metadata for Decisions scoring and Responses coaching feedback |
| Anthropic | Live conversation content when selected as a Hume language model |
| Other active Users in the same Demo instance | Your name and eligible transcripts for human evaluation; saved AI overall and capability scores after their own submission. Each reviewer’s human ratings, explanation, notes, and draft remain private to that reviewer |
| Google Gemini and search/tool providers | Relevant transcript, context, or tool requests when a tester selects a Hume variant using those services or a configured feature invokes a tool |
| Hosting and network providers, including Vercel | Requests, IP addresses, user agents, session and routing data, application responses, deployment logs, and other information needed to host and deliver the Demo |
| Database providers, including Neon | Account, session, practice, transcript, evidence, report, evaluation, deletion, and audit records stored in PostgreSQL |
| Private S3-compatible storage providers | Verified raw archives containing authorized microphone audio, assistant audio, and application frames, plus object metadata |
| Security, logging, and observability providers, including Sentry and Honeycomb where configured | Errors, logs, traces, metrics, request and correlation identifiers, configuration and provider details, and diagnostic payloads needed to secure, troubleshoot, and measure the Demo |
| Personnel and contractors | Personal information reasonably necessary for engineering, support, security, quality assurance, authorized research, model development, legal, and program administration |
| Professional advisers | Information reasonably necessary for legal, audit, accounting, insurance, security, or compliance services, subject to professional or contractual duties |
| Authorities and protected parties | Information Sparked reasonably believes must be disclosed to comply with law or lawful process, enforce agreements, investigate unlawful conduct, or protect rights, safety, and security |
| Transaction participants | Information reasonably necessary to evaluate or complete a merger, acquisition, financing, reorganization, bankruptcy, or sale of all or part of Sparked’s business or assets, subject to appropriate confidentiality protections |
Sparked does not disclose your identifiable recordings, transcripts, individual scores, or coaching reports to any third party except as described in this Policy. This restriction does not prevent you from sharing your own report or Sparked from making a disclosure required by law as described elsewhere in this Policy.
9. No sale, targeted advertising, or significant decisions
Sparked does not sell personal information for money or other valuable consideration and does not share personal information for cross-context behavioral advertising. Sparked does not knowingly sell or share personal information of anyone under 18. Any provider use of anonymized information described in Section 5 is subject to Sparked’s applicable account settings, contracts, and provider terms and is not a disclosure for advertising.
The Demo does not serve targeted advertisements or use third-party advertising cookies. The public marketing website may have different practices governed by its separate privacy notice.
Sparked uses automated systems to produce practice scores, coaching, and product-testing information, but Sparked does not use the Demo to make or facilitate decisions that produce legal or similarly significant effects concerning any person. Demo output is practice and product-testing information. It is not validated for, and must not be used to make or facilitate, any decision about hiring, retention, advancement, compensation, discipline, credentialing, credit, housing, insurance, healthcare, education admission, or access to essential goods or services.
You may contact Sparked to request information about a score or automated evaluation, identify a suspected error, or request human review. Human review does not guarantee that a score or report will change.
10. Retention
Sparked retains personal information described in this Policy for up to 36 months, measured from your last Demo activity, and uses it during that period for the purposes described in Sections 6 and 7.
Sparked may retain de-identified information, aggregate findings, derived analytics, evaluation methods, and model artifacts that cannot identify you indefinitely.
Sparked may retain a limited record beyond 36 months when necessary to document consent; comply with law or lawful process; investigate fraud, abuse, or a security incident; enforce agreements; or establish, exercise, or defend legal claims. Sparked will limit use during that additional period to the applicable purpose.
The period above governs Sparked-controlled records. It does not control Hume’s retained chat history, a Hume audio reconstruction, OpenAI’s abuse-prevention retention, or another provider-controlled copy. Provider-controlled retention follows the contracts, settings, legal duties, technical limits, and policies described in Section 5.
11. Your privacy rights and choices
Sparked offers the following request process to every Demo tester in the United States, whether or not a particular state privacy law applies:
- Access and know: request confirmation that Sparked processes your personal information and access to or a portable copy of personal information Sparked maintains about you, subject to legal and security restrictions.
- Correction: request correction of inaccurate account or personal information.
- Deletion: request deletion of personal information Sparked collected from you, subject to exceptions described below.
- Consent withdrawal: withdraw recording and improvement consent for future sessions. This ends future voice-session participation because those uses are essential to the Demo.
- Human review: ask for information about or human review of an automated coaching score or report.
- Appeal: appeal Sparked’s refusal of a privacy-rights request by replying to the decision and explaining why you believe it should be reconsidered.
- Authorized agent: where permitted by law, use an authorized agent to make a request on your behalf.
- Non-discrimination: exercise applicable privacy rights without unlawful discrimination.
Submit a request by emailing kat@sparkedinnovations.ai, calling +1 949-826-1060, or writing to the address in Section 18. Please use the subject line “Privacy Request — Trellium Demo” when emailing.
Sparked must verify that a request concerns you before disclosing, correcting, or deleting information. Sparked may ask you to respond from your provisioned email address or provide information that can be matched to account records. Sparked will use verification information only to process the request and prevent fraud. For an authorized agent, Sparked may request proof of the agent’s authority and may ask you to verify your identity or directly confirm authorization unless law provides otherwise.
Sparked will respond within the period required by applicable law and ordinarily aims to respond within 45 days. If Sparked needs an extension, it will provide notice when required. Sparked does not charge for reasonable requests but may deny or charge a legally permitted fee for requests that are manifestly unfounded, excessive, repetitive, fraudulent, or technically infeasible.
A deletion request may require Sparked to deactivate your account and will remove eligible identifiable records from active Sparked-controlled systems. Sparked may deny or limit deletion where information is necessary to provide a service you requested, secure the Demo, debug errors, exercise legal rights, comply with law, complete internal uses aligned with your relationship and consent, or satisfy another exception permitted by law. Deleted information may remain in provider systems as described in Section 5. Deletion cannot reverse prior use embodied only in de-identified data, aggregate statistics, evaluation methods, or non-identifying model artifacts.
12. State-specific disclosures
This Section supplements the rest of this Policy for residents of the states named below. Sparked voluntarily makes the rights process in Section 11 available to every U.S. tester, whether or not a state law applies. Where a state law grants a right or imposes a duty beyond that process, this Section describes it.
Every subsection below uses the same structure: Scope, Your rights, Sensitive data, Consent, Appeal, and How to submit.
12.1 California
Scope. The California Consumer Privacy Act, as amended by the California Privacy Rights Act (“CCPA”), applies to California residents to the extent it covers Sparked or a particular processing activity.
The table below describes categories of personal information Sparked has collected or expects to collect through the Demo during the preceding 12 months. Category labels correspond to California law.
| California category | Collected | Examples from the Demo | Disclosed for a business purpose | Sold or shared |
|---|---|---|---|---|
| A. Identifiers | Yes | Name, email, account and user IDs, IP address, session and chat identifiers | Other active Users in the same instance (name); Hume and AI providers; hosting, database, security, support, and observability providers | No |
| B. California Customer Records information | Yes | Name, contact information, account and authentication records | Account, hosting, database, security, and support providers | No |
| C. Protected classification characteristics | Possible if voluntarily revealed | Age, disability, race or ethnicity, sex or gender, religion, or other characteristics mentioned in a session or support request; Sparked does not request them | Voice, AI, storage, support, and infrastructure providers when included in submitted content | No |
| D. Commercial information | No | The Demo does not process purchases or payment cards | Not applicable | No |
| E. Biometric information used for identification | No | Sparked does not create voiceprints or process voice data to uniquely identify or authenticate testers | Not applicable | No |
| F. Internet or other electronic network activity | Yes | Usage, routes, interactions, browser and device data, event sequences, logs, cookies, errors, and diagnostics | Hosting, voice, AI, security, support, and observability providers | No |
| G. Geolocation data | No | Sparked does not derive or infer location from IP address and does not request precise geolocation | Not applicable | No |
| H. Sensory information | Yes | Microphone and assistant audio, transcripts, and selected application frames | Other active Users in the same instance (eligible transcripts); Hume and AI providers, private storage, authorized support and research personnel | No |
| I. Professional or employment-related information | Possible | Role, specialty, or professional context volunteered during a session | Voice, AI, storage, support, and infrastructure providers when included in submitted content | No |
| J. Nonpublic education information | No | Sparked does not request student records | Not applicable | No |
| K. Inferences | Yes | Expression predictions, communication metrics, annotations, scores, strengths, improvements, and coaching recommendations | Other active Users in the same instance (saved AI scores after submission); Hume, OpenAI, authorized personnel, and infrastructure providers | No |
| L. Sensitive Personal Information | Limited | Account login credentials; session content that may voluntarily reveal health, racial or ethnic origin, religious or philosophical beliefs, union membership, sexual orientation, or other sensitive information | Providers processing the account or content for the disclosed Demo purposes | No |
The sources of these categories are you; your browser, device, and network; Hume and other AI providers; infrastructure and security providers; and Sparked’s own calculations and inferences. Sparked uses these categories for the business and commercial purposes in Section 6. Sparked disclosed Categories A, B, C, F, H, I, K, and L for those business purposes during the preceding 12 months, to the extent the category was present, and did not sell or share any category for cross-context behavioral advertising.
Your rights. Subject to exceptions and verification, California residents may know the categories and specific pieces of personal information collected, the sources, purposes, and categories of recipients; request deletion; request correction; opt out of the sale or sharing of personal information; limit qualifying uses and disclosures of Sensitive Personal Information; access information about qualifying automated decisionmaking and opt out when applicable; and receive equal service and pricing without retaliation for exercising CCPA rights.
Sparked does not sell or share personal information for cross-context behavioral advertising, so an opt-out request will be recorded but does not change those practices. Sparked does not use automated decisionmaking to make a significant decision about a tester.
Sensitive data. Sparked uses Sensitive Personal Information only for the disclosed account, security, Demo operation, session analysis, support, authorized research and model-improvement, and legal purposes. Sparked does not use it to infer protected characteristics for advertising or to make significant decisions, and does not use voice data as a biometric identifier. Sparked does not presently use or disclose Sensitive Personal Information for a purpose that requires a Limit the Use of My Sensitive Personal Information link. If that changes, Sparked will provide the required notice and choice. You may still submit a limitation request under Section 11.
Consent. By accessing or using the Demo you consent to the recording, analysis, human review, and research and model-development uses described in this Policy and the Terms.
Appeal. Reply to Sparked’s decision explaining why you believe it should be reconsidered. Sparked will respond as required by law.
How to submit. Use any method in Section 11 with the subject line “Privacy Request — Trellium Demo.” You may use an authorized agent subject to the verification requirements in Section 11. For general information about California privacy rights, visit the California Attorney General’s CCPA page.
California “Shine the Light.” California Civil Code Section 1798.83 permits eligible residents to request information about disclosures of personal information to third parties for their direct-marketing purposes. Sparked does not disclose Demo personal information to third parties for their own direct marketing.
12.2 Illinois
Scope. The Illinois Biometric Information Privacy Act (“BIPA”) regulates biometric identifiers, including a voiceprint, and biometric information derived from them.
Sparked does not create or collect a voiceprint. Sparked records and analyzes voice to transcribe speech, measure probabilistic vocal-expression signals, and generate coaching. Sparked does not use voice data to identify or authenticate any individual, does not compare a recording against an identity database, and does not generate a template intended to recognize a person by voice. Speech, expression, and conversation models trained or evaluated on session audio are built to understand speech and conversational dynamics generally, not to identify any person.
Your rights. Illinois residents may exercise every right in Section 11. In addition, if any information Sparked holds were determined to be a biometric identifier or biometric information, Sparked would not sell, lease, trade, or otherwise profit from it, and would disclose it only with consent, to complete a transaction you requested, as required by law, or under a valid warrant or subpoena.
Sensitive data. Voice recordings, transcripts, and expression signals are treated as sensitive regardless of whether they meet the BIPA definition. Access is limited to personnel, contractors, and providers with a business need.
Retention and destruction. To the extent any Demo information were treated as a biometric identifier or biometric information, Sparked’s written retention schedule is the one stated in Section 10: up to 36 months from your last Demo activity, after which Sparked will destroy the information, except where a longer period is required to comply with law or lawful process, or to establish, exercise, or defend a legal claim. De-identified and non-identifying artifacts are not biometric identifiers and are retained as stated in Section 10.
Consent. By accessing or using the Demo you consent to the recording, transcription, expression analysis, human review, and research and model-development uses described in this Policy and the Terms. If Sparked determines that a written release is required for any activity, Sparked will obtain it before conducting that activity.
Appeal. Reply to Sparked’s decision explaining why you believe it should be reconsidered.
How to submit. Use any method in Section 11.
12.3 Texas
Scope. Two Texas laws may apply: the Capture or Use of Biometric Identifier Act (“CUBI”), which regulates capture of a voiceprint for a commercial purpose, and the Texas Data Privacy and Security Act (“TDPSA”).
Sparked does not capture a voiceprint and does not use voice data to identify or authenticate any individual, as described in Section 12.2.
Your rights. Texas residents may confirm whether Sparked processes their personal data and access it; correct inaccuracies; delete personal data; obtain a portable copy; and opt out of targeted advertising, sale of personal data, or profiling in furtherance of decisions producing legal or similarly significant effects. Sparked does not engage in any of those three activities for the Demo, so an opt-out will be recorded but does not change current practices.
Sensitive data. Session content may reveal sensitive data as the TDPSA defines it, including information about health, racial or ethnic origin, or religious beliefs. Sparked processes sensitive data only for the disclosed Demo, support, research, model-improvement, security, and legal purposes.
Consent. The TDPSA requires consent before processing sensitive data. By accessing or using the Demo you consent to Sparked’s processing of the session content and expression signals described in this Policy, including for research and model development.
Appeal. Reply to Sparked’s decision. If Sparked denies your appeal, you may contact the Texas Attorney General at texasattorneygeneral.gov.
How to submit. Use any method in Section 11.
12.4 Washington
Scope. The Washington My Health My Data Act (“MHMDA”) regulates consumer health data, which is defined broadly and may include inferences about a consumer’s physical or mental health drawn from other data. Probabilistic vocal-expression signals and session content that touches on health may fall within that definition.
Your rights. Washington residents may confirm whether Sparked collects, shares, or sells their consumer health data; access that data, including a list of all third parties and affiliates with whom Sparked has shared or sold it; withdraw consent to collection and sharing; and request deletion of consumer health data from Sparked’s records, including from archived and backup systems as required by the MHMDA.
Sensitive data. Sparked treats voice recordings, transcripts, and expression signals as sensitive. Sparked does not use them to infer health conditions for advertising, and does not disclose them for advertising purposes.
Consent. By accessing or using the Demo you consent to the collection, processing, and sharing of consumer health data as described in this Policy, including the recording, expression analysis, human review, and research and model-development uses. You may withdraw consent for future sessions under Section 11; because recording and analysis are essential to the Demo, withdrawal ends future voice-session participation.
No sale. Sparked does not sell consumer health data. The MHMDA requires a separate signed authorization for any such sale, and Sparked does not seek one.
Appeal. Reply to Sparked’s decision explaining why you believe it should be reconsidered. You may also contact the Washington Attorney General.
How to submit. Use any method in Section 11 with the subject line “Privacy Request — Trellium Demo.”
12.5 Nevada
Scope. Nevada Senate Bill 370 regulates consumer health data, and Nevada Revised Statutes Chapter 603A gives consumers a right to opt out of the sale of covered information.
Your rights. Nevada residents may confirm whether Sparked collects, shares, or sells their consumer health data; access that data and the list of third parties with whom it has been shared or sold; withdraw consent to its collection and sharing; request deletion; and direct Sparked not to sell covered information.
Sensitive data. Sparked treats voice recordings, transcripts, and expression signals as sensitive and processes them only for the purposes described in this Policy.
Consent. By accessing or using the Demo you consent to the collection, processing, and sharing of consumer health data as described in this Policy.
No sale. Sparked does not sell consumer health data or covered information. An opt-out request will be recorded but does not change current practices.
Appeal. Reply to Sparked’s decision explaining why you believe it should be reconsidered.
How to submit. Use any method in Section 11.
12.6 Other states with comprehensive privacy laws
This subsection applies to residents of Colorado, Connecticut, Virginia, Utah, Oregon, Montana, Florida, Delaware, Iowa, Nebraska, New Hampshire, New Jersey, Tennessee, Minnesota, Maryland, Indiana, Kentucky, and Rhode Island. Each state’s law differs in detail, effective date, and available exemptions. Sparked applies the process below uniformly to residents of every state named here, and will honor any additional right a particular state’s law grants.
Scope. These laws apply to personal data Sparked processes about a resident of the named state, subject to that state’s thresholds and exemptions.
Your rights. Subject to verification and applicable exceptions, you may confirm whether Sparked processes your personal data and access it; correct inaccuracies; delete personal data you provided or Sparked obtained; obtain a portable copy in a readily usable format; and opt out of targeted advertising, the sale of personal data, and profiling in furtherance of decisions producing legal or similarly significant effects. Sparked does not conduct targeted advertising, sell personal data, or profile testers for significant decisions in the Demo, so an opt-out will be recorded but does not change current practices. Where a state recognizes a universal opt-out mechanism such as Global Privacy Control, Sparked will honor it for any practice it covers, as described in Section 13.
Residents of Maryland should note that their state law restricts the sale of sensitive data outright and limits collection to what is reasonably necessary for the requested product or service. Residents of Minnesota may additionally request a list of the specific third parties to which Sparked has disclosed their personal data. Residents of Oregon may request a list of specific third parties to which Sparked has disclosed any personal data. Sparked will honor those requests through the same process.
Sensitive data. Session content and expression signals may constitute sensitive data under these laws. Sparked processes them only for the disclosed Demo, support, research, model-improvement, security, and legal purposes, and does not use them for advertising or significant decisions.
Consent. Most of these states require consent before processing sensitive data. By accessing or using the Demo you consent to Sparked’s processing of the session content and expression signals described in this Policy, including for research and model development. You may withdraw consent for future sessions under Section 11.
Appeal. If Sparked declines your request, reply to the decision explaining why you believe it should be reconsidered. Sparked will respond within the period the applicable state law requires and will tell you how to contact your state Attorney General if the appeal is denied.
How to submit. Use any method in Section 11 with the subject line “Privacy Request — Trellium Demo,” and identify your state of residence so Sparked can apply the correct deadlines.
13. Do Not Track and Global Privacy Control
Some browsers offer Do Not Track (“DNT”) signals. There is no uniform standard that requires a particular response to DNT, and the Demo does not change its operation in response to DNT. The Demo uses only the authentication, security, storage, performance, and operational technologies described in this Policy, not cross-site advertising trackers.
Global Privacy Control (“GPC”) may communicate a request to opt out of sale or cross-context behavioral-advertising sharing. Sparked does not engage in those activities for the Demo, so a GPC signal does not require a change to the Demo’s current processing. If Sparked begins a practice subject to GPC, Sparked will honor the signal as required by applicable law.
Neither DNT nor GPC withdraws your consent to session recording or product and model improvement. To withdraw that consent for future sessions, use the process in Section 11.
14. Children’s privacy
The Demo is limited to adults aged 18 or older. Sparked does not knowingly collect personal information through the Demo from anyone under 18. If you believe a minor has provided personal information, contact Sparked immediately. After verification, Sparked will deactivate the account and delete the information as required by applicable law.
15. Security
Sparked uses measures designed to protect Demo information, including authenticated user-owned records, encrypted network transport, private object storage, scoped access controls, credential redaction, integrity hashes, security headers, rate limiting, and audit or deletion workflows. Raw archives are not separately encrypted by Sparked at the application layer; their confidentiality depends on private storage controls, transport encryption, provider safeguards, and authorized access.
No security measure or transmission is infallible. Sparked cannot guarantee absolute security, that every diagnostic field will be redacted, or that a provider will never experience an incident. Use a trusted device and network, keep your password confidential, use Lock when finished, and promptly report suspected unauthorized access to Sparked.
16. U.S. processing and business transfers
The Demo is offered only in the United States. Sparked and its providers process and store information in the United States, although a provider may operate personnel or infrastructure in other locations under its contractual and legal obligations.
If Sparked is involved in a merger, acquisition, financing, reorganization, bankruptcy, or sale of all or part of its relevant business or assets, personal information may be reviewed or transferred as part of the transaction. Sparked will require a recipient to handle personal information consistently with this Policy unless and until it provides a new notice and obtains any consent required by law.
17. Changes to this Policy
Sparked may update this Policy to reflect changes in the Demo, providers, practices, or law. The document version and effective date at the top identify the current version. Sparked may provide notice through the Demo, by email, or by another reasonable method.
If a change materially expands the categories collected, the purposes for which identifiable information is used, or the parties’ rights and obligations, Sparked will increment the major document version, present the new version, and obtain additional consent where required. A new version will not retroactively authorize an incompatible use of information collected under an earlier version without any consent required by law.
18. Contact us
Questions, consent withdrawals, appeals, complaints, and privacy requests may be sent to:
Sparked Innovations.AI 8 The Green Ste B Dover, DE 19901 United States Email: kat@sparkedinnovations.ai Telephone: +1 949-826-1060
For a privacy request, use the email subject “Privacy Request — Trellium Demo” and identify the provisioned email address associated with your account and your state of residence.