How We Store a Soul: Designing Data Systems That Respect Emotional Privacy

by Vivian Lei and Chuyu Duan

At PowerYou, our mission is to create an emotionally intelligent AI guide that supports healing, growth, and transformation. For our users to trust Kris with their most vulnerable thoughts and emotions, our systems must do more than just store data — they must protect it like something sacred.

This post dives into the architecture, encryption protocols, and world-class edge-case handling that power PowerYou’s commitment to emotional privacy. Because when a user shares their inner world with us, they’re not giving us content to monetize. They’re entrusting us with something closer to a soul.

We Don’t Sell Data. Period.

Let’s start with the business model.

PowerYou don’t sell user data because our business model is not ad-based. We don’t track emotional vulnerabilities to serve targeted content. Our users are not the product - they’re the customer. This foundational stance informs every technical and ethical decision we make.

Unlike many social media platforms or entertainment-based apps like TikTok, Instagram, or Snapchat, we don’t collect emotional insights to refine engagement algorithms or fuel advertising profiles. Our incentive is not to hook users, but to help them heal and grow.

Because we serve people directly, we design for deep trust, not shallow engagement. Our infrastructure is built not just to meet compliance requirements, but to exceed the emotional expectations of someone trusting an AI with their inner world.

The Core System: Contextual Memory Containers

At the heart of Kris’s intelligence is a system we call Contextual Memory Containers.

Each container holds:

• A slice of emotional context (e.g. grief, burnout, healing from trauma)

• A user-defined intention or journey (e.g. "Build Self-Trust")

• Relevant emotion logs, voice sessions, insights, and reflections

  
  

These containers are separately encrypted and user-scoped, meaning they’re never blended across users, even at the analytics level. A user’s healing journey is theirs alone.

Encryption by Default

We employ a multi-layered encryption model across all data operations:

• At rest: All data is encrypted using AES-256 via AWS KMS. This includes all DynamoDB tables where chat history, emotion logs, and memory container metadata are stored.

  
  

• In transit: We enforce TLS 1.2+ across all API interactions, including internal services and LLM payloads.

  
  

• Field-level encryption: For especially sensitive fields, we use advanced field-level encryption and user-scoped keying to ensure that even if unauthorized access occurs, sensitive emotional data remains unreadable.

  
  

• Tokenization: User identifiers are tokenized at the ingestion layer so internal systems never operate on raw PII. This separates user identity from emotional data, preserving anonymity in internal observability tools.

  
  

Zero Third-Party Sharing

We do not share user data with:

• Advertisers

• Brokers

• Social media integrations

• Data enrichment vendors

  
  

Every third-party vendor we use is bound by strict DPA agreements. No vendor is permitted to retain or use PowerYou data for training, profiling, or resale. We routinely audit third-party activity logs and minimize data exposure to the absolute minimum necessary for functionality.

Edge Case Handling: Designed for the Worst Day

World-class emotional safety demands preparation for edge cases. Here’s how we approach it:

1. Partial Data Loss

Every critical interaction is redundantly stored and validated across multiple layers to ensure durability, accuracy, and real-time recovery if a session is interrupted.

If any corruption or schema conflict is detected, the affected record is quarantined and does not affect the user session.

2. Session Handoff Errors

If a user closes the app mid-reflection or loses network:

• The session state is checkpointed locally

• Redis-based cache stores recent interactions with TTL fallback

• We support seamless reconnection and continuity, so users never lose their progress - even if interrupted mid-session.

  
  

3. Unauthorized Access Attempt

We deploy behavior-based anomaly detection:

• We continuously monitor for signs of unauthorized access using behavioral patterns, with real-time intervention and identity re-verification when necessary.

• All access attempts are logged and monitored with automated alerting through AWS CloudTrail and GuardDuty

If suspicious behavior is detected, the user’s account is auto-locked and requires a secure identity verification flow to unlock.

4. Human-In-The-Loop Logging Exceptions

Some sessions may be flagged for optional human review if safety risk is suspected. These are:

• Not visible to any third-party reviewers

• Only reviewed by an internal licensed responder if the user has opted into safety features

• Logged separately from standard memory containers with a distinct access path

Respect Beyond Regulation

Our approach isn’t just about compliance with regulation. It’s about respect.

We assume our users are trusting us with information they may not even share with their therapist or loved ones. That deserves the highest level of technical care and ethical restraint.

When you share intimate details of your inner world, you are not starting a funnel. You are starting a relationship. And that relationship lives inside a system designed to keep your vulnerability safe.

Final Thought

PowerYou is not just building emotional AI. We’re building emotional infrastructure — systems that hold memory, identity, growth, and pain with dignity.

The future of AI won’t just be about intelligence. It will be about trust.

And trust begins with how we store a soul.