Lattice Privacy Policy

This is the Privacy Policy for the invited Lattice alpha. Questions or data requests: david@latticelanguages.com.

Effective date: pending publication

Version: 1 (matches in-app consent version 1)

Who we are: Lattice Language Learning LLC, a Washington limited liability company.

1. What Lattice is, and what this alpha is

Lattice is a language-learning app (Japanese first) built around a persistent model of what you know and are learning. Right now Lattice is in a private alpha: roughly 40 invited testers, by invitation only. It is run by one person (David Luongo). This policy is written to be honest about what that means:

Some things are handled manually (deletion and export requests, for example) rather than through self-serve buttons.
The product changes fast. If what we collect or how we process it changes materially, we will update this policy, bump the consent version, and ask you to re-consent before the affected feature processes your data.

This policy tells you what we collect, why, who processes it, how long we keep it, and how to get it deleted or exported.

2. What we collect

2.1 Account data

Your email address. This is the only account identifier; there are no passwords (login is by emailed magic link). Your email is the only place your real-world identity attaches to your learning data.
Login session records (an opaque session token and metadata such as when the session was created).

2.2 Learning data (the learner model)

Lattice's core is a ledger of your learning activity. This includes:

Every word or item you encounter, and the app's evolving estimate of whether you know it
Your learner profile (display name if you set one, estimated level)
Notes and journal entries you write in the app (free text)
Session history, review results, progress and streak-style records
Words and short context snippets transcribed from your voice tutoring sessions (text, not audio)

This is the point of the product: the app remembers what you are learning so it can teach you better. It is also the most personal data we hold, and it is scoped to your account.

2.3 Voice audio

If you use the voice tutor, your voice audio is captured by your device and sent to speech-processing providers (listed in §5) to transcribe what you said and generate the tutor's reply. Lattice's own servers store the resulting transcripts (text), not the raw audio. Voice audio is subject to the specific biometric consent described in §6, and we will not process your voice without it.

2.4 Diagnostics

Error and crash reports: when something breaks, the app sends a diagnostic event (the page or endpoint involved, a stack trace, an error message, your browser or device user agent). Error messages can occasionally contain fragments of what was on screen.
Usage events: app opens, which surfaces you visit, session ratings and bug reports you choose to submit. During the alpha we use these to measure whether the product works (bug counts, session quality ratings, retention) — that measurement plan is the reason the alpha exists.
Generated images: the app sometimes generates visual aids from session context; the generated image files are stored linked to the session that produced them.

2.5 What we do NOT collect

No ad trackers, no third-party analytics SDKs, no advertising identifiers.
No payment data (the alpha is free).
No contacts, location, photos, or anything else from your device beyond the microphone (voice tutor only, with consent) and what you type.

3. Why we process it (purposes)

The app asks your consent for specific, named purposes. These match the purposes recorded in our consent system, version 1:

Purpose	What it covers
terms	Agreeing to the Terms of Service and this Privacy Policy
data_processing	Processing of your learning activity to run the tutor — the learner model in §2.2
voice_biometric	Processing of voice recordings (a biometric identifier) to power speech-based tutoring — see §6

The first two are required to use the app at all. Voice consent is separate and optional: you can use Lattice without ever granting it, and the voice tutor simply stays off.

We additionally process diagnostics (§2.4) on the basis of running and improving the alpha; during the alpha, participating in basic measurement (bug reports, ratings, usage events) is part of what testers sign up for, and the tester agreement covers it.

4. Where your data lives

Production data is stored on Google Cloud Platform in the United States (us-west1, Oregon): a managed PostgreSQL database and an application server. The database has point-in-time recovery and automated backups, and the server disk has daily snapshots. Backup copies age out on Google's standard schedules; a deletion request removes you from the live systems immediately and from backups as they expire (see §8).

5. Who else processes your data (our vendors)

We do not sell your data to anyone (§10). We do use a small set of processing vendors to run the product. What each one sees:

Vendor	What flows to them	Notes
Google Cloud (hosting, US)	Everything in §2, at rest and in transit on our infrastructure	Our cloud provider; data stays in the US region
OpenAI (realtime voice tutors; image generation)	Raw voice audio during realtime voice sessions; image-generation prompts	Per OpenAI's API policy, API data is not used to train their models by default, but OpenAI retains API data for up to about 30 days for abuse monitoring. We are requesting a zero-data-retention arrangement; until granted, assume the 30-day window applies to voice audio sent to OpenAI.
Deepgram (speech-to-text)	Raw voice audio, for transcription	We set the opt-out flag on every request that excludes your audio from Deepgram's model-improvement program; audio is retained only as needed to process the request
Anthropic (chat tutor)	Your chat messages and tutor context (text)	API data is not used for training by default; standard API retention applies
Cerebras (language model for the composed voice stack)	Transcripts of what you said, as text prompts	Used when the composed voice stack is active
Cartesia (text-to-speech)	The tutor's reply text only (which can reference your learning context); never your voice	Used when the composed voice stack is active
Resend (login emails)	Your email address	Transactional login emails only

6. Voice consent (biometric data)

We treat voice recordings as potentially containing a biometric identifier (a voiceprint) and hold ourselves to the standard of biometric-privacy laws such as Illinois BIPA, which require informed, specific, written consent BEFORE any capture. Concretely:

What is collected: audio of your voice, captured by your device microphone during voice tutoring sessions only. The microphone is not on at any other time.
Why: to transcribe your speech, evaluate your spoken Japanese, and generate the tutor's spoken replies. That is the whole purpose. We do not use your voice to identify you, to build a voiceprint for authentication, or for anything unrelated to tutoring.
Consent before capture: the app will not open a voice session until you have granted the voice_biometric consent, recorded with a timestamp and version. This is enforced in the server, not just the UI: the voice endpoint refuses without it.
Retention and destruction: Lattice's servers keep transcripts (text), not audio. Raw audio exists (a) transiently in your device's audio buffers, and (b) at the speech vendors in §5 for as long as their processing and stated retention windows require (up to about 30 days at OpenAI; processing-only at Deepgram).
Revocation: you can revoke voice consent at any time by emailing david@latticelanguages.com. Revocation stops all further voice capture immediately (the server gate closes). It does not undo processing that already happened, but you can pair revocation with a deletion request (§8) to remove your transcripts. During the alpha, revocation is by email; an in-app settings control is planned.

7. How long we keep things (retention)

Data	Retention
Account email + account record	For the life of your account; deleted on account deletion
Login magic links	15 minutes, single use
Login sessions	30 days
Learning data (ledger, notes, journal, transcripts)	For the life of your account; deleted on account deletion
Voice audio	Not stored by Lattice; vendor windows per §5 and §6
Diagnostics / error logs	Kept as needed to debug the alpha and periodically cleared; not retained indefinitely
Generated visual-aid images	Session-linked; swept on deletion requests
Safety-critical audit events	See below
Backups	Deleted data ages out of backups on the standard backup expiry schedule (§4)

Safety-critical events: our internal audit log records event categories (not message content, not emails). Most audit rows follow a standard retention period, but records of crisis-safety events (for example, a safety intervention during a tutoring session) are retained permanently as a deliberate policy, including through account deletion, in a legal-hold-style posture.

8. Deletion and export (your data, your call)

During the alpha, deletion and export are manual but real:

Email david@latticelanguages.com from your account email and ask for deletion, export, or both.
Deletion removes your account, email, learning data, transcripts, notes, journal, session records, generated images, and your diagnostic-log entries from the live systems. Safety-critical audit events are exempt (§7). Backup copies expire on the backup schedule rather than instantly (§4).
Export gets you a machine-readable copy of your learning data.
Target turnaround during the alpha: within 14 days, usually much faster.

We are telling you plainly: there is no self-serve delete button yet. There is one email address, read by the one person who runs this, and a documented erasure map covering every store your data touches.

9. What we share, and with whom

The vendors in §5, strictly to run the product.
Nobody else, with the standard exceptions: if the law compels us, to protect someone's safety, or as part of a future corporate change (in which case this policy's promises travel with the data).

10. No ads, no sale of data

We do not sell your personal data. Not to advertisers, not to data brokers, not to anyone.
We do not run ads and there is no advertising tracking in the product.
We do not use your data to train our own or anyone else's foundation models, and our LLM vendors' API terms exclude training use by default (§5).

11. Who can use the alpha

The alpha is for invited testers aged 18 or older. It is not directed at children and we do not knowingly collect data from anyone under 18. If you believe someone under 18 has an account, contact us and we will delete it.

12. Your rights

Wherever you are, the practical rights are the ones in §8: access, export, deletion, and consent revocation, all via david@latticelanguages.com.

13. Changes to this policy

When we change this policy in a way that matters, we will: update the version number, note the effective date, ask for fresh consent in the app before affected processing continues (the consent system versions every grant), and email testers about material changes. The current version is always at https://api.latticelanguages.com/privacy.

14. Contact

Questions, deletion or export requests, consent revocation, or anything else:
david@latticelanguages.com