Security & Compliance

Data inventory

Data categories, flows, and retention owners across the Perceive8 platform.

Data inventory

Version: 1.0
Last Updated: 2026-03-07
Owner: Privacy & Compliance Team
Classification: Internal — Confidential
Related Docs: Data retention · Consent model · Subprocessors · Privacy requests

1. Overview

This document provides a comprehensive inventory of all personal and sensitive data collected, processed, and stored by the Perceive8 meeting intelligence platform. It is organized by feature area and includes data classification, storage location, processing parties, and cross-border transfer implications.

Data Classification Levels

Level	Label	Description
L1	Public	Non-sensitive, publicly available information
L2	Internal	Internal operational data, low sensitivity
L3	Confidential	Business-sensitive data, PII, requires access controls
L4	Restricted	Highly sensitive data — biometrics, health data, legal recordings

2. Data Inventory by Feature Area

2.1 Authentication & User Management

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
`external_id` (Supabase auth UUID)	Account identification — Contract performance	PostgreSQL (Supabase)	Supabase (internal)	L3 — Confidential	Yes (pseudonymous identifier)	Supabase cloud region (US)
Email address	Authentication, account recovery — Contract performance	Supabase Auth	Supabase	L3 — Confidential	Yes	Supabase cloud region (US)
Password hash	Authentication — Contract performance	Supabase Auth	Supabase	L4 — Restricted	No (derived)	Supabase cloud region (US)
Session tokens / JWTs	Session management — Contract performance	In-memory / client-side	Internal	L3 — Confidential	Yes (linked to user)	Client device location

users table:

Stores external_id linking to Supabase Auth
Minimal PII stored in application database by design
Auth metadata (email, password) managed by Supabase Auth service

2.2 Audio Processing

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Raw audio files	Meeting analysis — Consent	MinIO (S3-compatible object storage)	Internal	L4 — Restricted	Yes (voice recordings)	Self-hosted / deployment region
Enhanced audio files	Improved transcription quality — Consent	MinIO (S3-compatible object storage)	Internal	L4 — Restricted	Yes (voice recordings)	Self-hosted / deployment region
`original_path`	File reference — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to audio)	Supabase cloud region
`enhanced_path`	File reference — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to audio)	Supabase cloud region
Audio metadata (duration, format, sample rate)	Processing optimization — Legitimate interest	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region

audio_files table:

Contains paths to raw and enhanced audio in object storage
Audio files are the most sensitive data class — contain voice biometrics and spoken content
Linked to analyses via foreign key

2.3 Transcription

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Audio sent to OpenAI Whisper	Speech-to-text transcription — Consent	Transient (API call)	OpenAI (third-party)	L4 — Restricted	Yes	US (OpenAI servers)
Audio sent to Replicate	Diarized transcription — Consent	Transient (API call)	Replicate (third-party)	L4 — Restricted	Yes	US (Replicate servers)
Transcript text (`text`)	Core product feature — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (verbatim speech)	Supabase cloud region
`speaker_label`	Speaker attribution — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (identifies individuals)	Supabase cloud region
`language`	Transcription configuration — Contract performance	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region

transcript_segments table:

Contains verbatim speech content with timestamps
Speaker labels link to speakers table
Text may contain any PII spoken during the meeting (names, addresses, financial data, health information)

⚠️ Special Note: Transcript text is unpredictable in content. It may contain special category data (health, legal, financial) depending on the meeting subject matter. This must be treated as L4 — Restricted by default.

2.4 Speaker Analysis & Diarization

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Audio sent to Pyannote.ai	Speaker diarization & voiceprint — Consent	Transient (API call)	Pyannote.ai (third-party)	L4 — Restricted	Yes (biometric)	EU (Pyannote servers)
Speaker name (`name`)	Speaker identification — Consent	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes	Supabase cloud region
Diarization segments	Speaker turn tracking — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (speaker identification)	Supabase cloud region
Voiceprint embeddings	Speaker re-identification — Explicit consent	PostgreSQL (Supabase)	Internal	L4 — Restricted	Yes (biometric data)	Supabase cloud region
Speaker voice embeddings (vector)	Similarity matching — Explicit consent	ChromaDB	Internal	L4 — Restricted	Yes (biometric data)	Self-hosted / deployment region

speakers table:

Real names of meeting participants
Linked to voiceprint profiles for re-identification

voiceprint_profiles table:

Contains voice embeddings — classified as biometric data under GDPR Art. 9
Requires explicit consent for processing
Subject to enhanced protection requirements

diarization_segments table:

Speaker turn boundaries and identification data
Links audio segments to identified speakers

🔴 Biometric Data Warning: Voiceprint profiles constitute biometric data under GDPR, BIPA (Illinois), and similar regulations. Processing requires explicit consent and enhanced safeguards. See Consent Model.

2.5 Emotion & Prosody Analysis

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Audio sent to Hume AI	Emotion/prosody analysis — Consent	Transient (API call)	Hume AI (third-party)	L4 — Restricted	Yes	US (Hume AI servers)
`prosody_emotions`	Emotional tone analysis — Consent	PostgreSQL (Supabase)	Internal	L4 — Restricted	Yes (emotional state)	Supabase cloud region
`vocal_burst`	Non-speech vocal analysis — Consent	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (behavioral data)	Supabase cloud region

audio_intelligence_segments table (emotion fields):

Prosody emotions reveal emotional states of identified speakers
May constitute "inferences" about individuals under CCPA
Combined with speaker identification, this is highly sensitive profiling data

⚠️ Sensitive Processing: Emotion analysis combined with speaker identification constitutes profiling under GDPR Art. 22. Users must be clearly informed and given the right to object.

2.6 Content Intelligence (Sentiment, Entities, Topics, Moderation)

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Audio sent to AssemblyAI	Multi-modal analysis — Consent	Transient (API call)	AssemblyAI (third-party)	L4 — Restricted	Yes	US (AssemblyAI servers)
`sentiment_label`	Sentiment analysis — Consent	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (opinion attribution)	Supabase cloud region
`entities`	Named entity recognition — Consent	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (names, orgs, locations)	Supabase cloud region
`content_moderation`	Safety/compliance flagging — Legitimate interest	PostgreSQL (Supabase)	Internal	L3 — Confidential	Potentially	Supabase cloud region
`contextual_analysis`	Topic/context extraction — Consent	PostgreSQL (Supabase)	Internal	L3 — Confidential	Potentially	Supabase cloud region

audio_intelligence_segments table (content fields):

Entities may include PII extracted from speech (person names, organizations, locations, dates)
Content moderation flags may indicate sensitive topics discussed
Contextual analysis provides topic categorization

2.7 Conversations & Chat (RAG)

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Chat messages (`content`)	AI-assisted meeting analysis — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Potentially (user queries)	Supabase cloud region
Chat messages sent to OpenAI	RAG query processing — Consent	Transient (API call)	OpenAI (third-party)	L3 — Confidential	Potentially	US (OpenAI servers)
`tool_calls`	Function call metadata — Contract performance	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region
`artifacts`	Generated analysis artifacts — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Potentially	Supabase cloud region
Vector embeddings	Semantic search — Contract performance	ChromaDB	Internal + OpenAI (embedding generation)	L3 — Confidential	Yes (derived from transcripts)	Self-hosted (ChromaDB) / US (OpenAI API)

conversations table:

Links chat sessions to users
Contains conversation metadata

conversation_messages table:

Full chat history including user queries and AI responses
Tool calls may reference analysis data
Artifacts may contain generated summaries with PII

2.8 Scenarios & Reports

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Scenario definitions	Analysis templates — Contract performance	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region
`scenario_questions`	Analysis prompts — Contract performance	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region
`executive_summary`	Report output — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Potentially	Supabase cloud region
`answer` / `evidence`	Report findings — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (derived from transcripts)	Supabase cloud region

reports / report_findings tables:

Generated analysis reports containing summaries and evidence
Evidence text is extracted from transcripts and may contain PII
Reports may be shared or exported by users

2.9 Billing & Subscriptions

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
`stripe_customer_id`	Payment processing — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to user)	Supabase cloud region
`stripe_subscription_id`	Subscription management — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to user)	Supabase cloud region
`stripe_payment_id`	Payment tracking — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to user)	Supabase cloud region
Payment card details	Payment processing — Contract performance	Stripe only (not stored locally)	Stripe (third-party)	L4 — Restricted	Yes	US/EU (Stripe servers)
Usage minutes	Billing calculation — Contract performance	PostgreSQL (Supabase)	Internal	L2 — Internal	Yes (linked to user)	Supabase cloud region

subscriptions / usage_records / minute_packs tables:

Only Stripe reference IDs stored locally — no PCI data
Usage records track minutes consumed per analysis
Stripe handles all payment card data (PCI DSS compliant)

2.10 Real-Time Streaming

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
Real-time audio stream	Live transcription — Consent	Transient (WebSocket)	Internal + AI providers	L4 — Restricted	Yes (voice recordings)	Deployment region + provider regions
Stream session metadata	Session management — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to user)	Supabase cloud region
Stream alerts (`message`, `evidence_text`, `speaker_name`)	Real-time notifications — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes	Supabase cloud region

stream_sessions table:

Tracks active and historical streaming sessions
Links to user identity

stream_alerts table:

Contains alert messages with evidence text from live transcription
Speaker names directly identify individuals

2.11 API Access & Security

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
`key_prefix` / `key_hash`	API authentication — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to user)	Supabase cloud region
API key scopes	Authorization — Contract performance	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region
`client_id` / `client_secret_hash`	OAuth authentication — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to user)	Supabase cloud region
Webhook URLs / secrets	Event delivery — Contract performance	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes (linked to user)	Supabase cloud region

api_keys / oauth_clients / webhooks tables:

API keys stored as hashes only (prefix retained for identification)
OAuth client secrets stored as hashes
Webhook secrets used for payload signing

2.12 Audit & Compliance

Data Element	Purpose / Lawful Basis	Storage Location	Processor	Classification	PII	Cross-Border
`user_id`	Audit attribution — Legitimate interest / Legal obligation	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes	Supabase cloud region
`action`	Activity tracking — Legitimate interest	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region
`ip_address`	Security monitoring — Legitimate interest	PostgreSQL (Supabase)	Internal	L3 — Confidential	Yes	Supabase cloud region
Timestamp	Audit trail — Legal obligation	PostgreSQL (Supabase)	Internal	L2 — Internal	No	Supabase cloud region

audit_log table:

Immutable audit trail for compliance
IP addresses are PII under GDPR
Retention governed by legal/regulatory requirements

3. Cross-Border Data Transfer Summary

Destination	Provider(s)	Data Types	Transfer Mechanism	Status
United States	OpenAI, Replicate, Hume AI, AssemblyAI, Stripe	Audio, transcripts, embeddings, payment refs	Standard Contractual Clauses (SCCs) required	❌ SCCs not yet executed
European Union	Pyannote.ai	Audio, diarization data	Adequate (if EU-to-EU)	⚠️ Depends on deployment region
United States	Supabase	All database records	SCCs required (if users in EU)	❌ SCCs not yet executed

⚠️ Gap: No Standard Contractual Clauses (SCCs) or Transfer Impact Assessments (TIAs) have been executed with any AI provider. This is a prerequisite for processing EU personal data. See Subprocessors for DPA status.

4. Data Flow Diagram (Text)

User Device
    │
    ├── Audio Upload ──────────────────────┐
    │                                       ▼
    │                              ┌─────────────────┐
    │                              │   MinIO (S3)     │
    │                              │  Audio Storage   │
    │                              └────────┬────────┘
    │                                       │
    │                                       ▼
    │                              ┌─────────────────┐
    │                              │  Perceive8 API   │
    │                              │  (Processing)    │
    │                              └────────┬────────┘
    │                                       │
    │              ┌────────────────────────┼────────────────────────┐
    │              │            │           │           │            │
    │              ▼            ▼           ▼           ▼            ▼
    │        ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐ ┌──────────┐
    │        │  OpenAI   │ │Pyannote  │ │Replicate │ │Hume AI │ │AssemblyAI│
    │        │ Whisper   │ │  .ai     │ │          │ │        │ │          │
    │        └─────┬─────┘ └────┬─────┘ └────┬─────┘ └───┬────┘ └────┬─────┘
    │              │            │           │           │            │
    │              └────────────┴───────────┴───────────┴────────────┘
    │                                       │
    │                                       ▼
    │                              ┌─────────────────┐
    │                              │   PostgreSQL     │
    │                              │   (Supabase)     │
    │                              └────────┬────────┘
    │                                       │
    │                                       ▼
    │                              ┌─────────────────┐
    │   Chat / RAG ───────────────▶│    ChromaDB      │
    │        │                     │  (Embeddings)    │
    │        │                     └─────────────────┘
    │        │
    │        └──── OpenAI (Chat/Embeddings API)
    │
    └── Billing ──── Stripe (Payment Processing)

5. Implementation Status

Area	Status	Notes
Data inventory documentation	✅ Complete	This document
Data classification labels in code	❌ Not Implemented	No field-level classification metadata in database
Automated data discovery	❌ Not Implemented	Manual inventory only
Data flow monitoring	❌ Not Implemented	No runtime data flow tracking
Cross-border transfer safeguards	❌ Not Implemented	SCCs/TIAs needed for all US providers
PII detection in transcripts	❌ Not Implemented	No automated PII scanning of transcript content
Data minimization review	⚠️ Partial	Minimal user data stored; transcript content unbounded

6. Revision History

Version	Date	Author	Changes
1.0	2026-03-07	Privacy & Compliance Team	Initial data inventory

← Back to docs