This Too Shall Pass

How it works
Out roadmap
The user UI
Caregiver/Sysadmin UI

Inspiration

Grief is universal, but support isn’t.
Breakups, loss, and emotional crises often leave people isolated—blocked by stigma, cost, or lack of awareness.

We were driven by a simple question:
What if emotional support didn’t disappear when someone does?

At the same time, we recognized a critical boundary, AI should never replace real human relationships.
So we designed a system that acts as a temporary emotional bridge, not a permanent substitute.

What it does

this too shall pass is an AI-driven therapeutic support platform that helps users process grief, emotional distress, and crisis safely.

We translate the five stages of grief into adaptive AI behavior:

Venting (Denial & Anger):
The AI listens, validates, and encourages emotional expression.
Cloning (Bargaining & Depression):
The system reconstructs supportive elements (speech style, personality traits, shared memories) of the lost person to provide familiarity, while maintaining therapeutic boundaries.
Transitioning (Acceptance):
The AI gradually reduces mimicry and guides users toward independence, new habits, and real-world connections.

Safety & Ethical Design

We explicitly define what AI should not do:

Replace human relationships
Create emotional dependency
Provide clinical diagnosis

Instead, we built:

Stepped-care model: AI → Volunteers → Professionals
Crisis detection system: Real-time identification of self-harm risk
No-self-harm protocol: Encourages seeking immediate help
Memory modulation: Gradually tones down emotional reliance on the AI

Evidence-Based Approach

Our system is grounded in psychological theory and measurement.

We model emotional progression as a dynamic function over time:

$$ Recovery(t) = f(Emotional\ Expression, Memory\ Exposure, Support\ Level) $$

We track distress using the Impact of Event Scale (IES) and adapt intervention intensity accordingly:

$$ If\ IES_{score} > \theta \rightarrow Escalate\ to\ Human\ Support $$

This ensures that the system is not static, but responsive to user state.

How we built it

AI Engine: Google Gemini (context-aware therapeutic responses)
Memory Extraction: Structured user inputs into memories, habits, routines, and traits
Backend: Firebase (Authentication, Firestore database, role-based access)
Crisis Analyzer: Detects high-risk signals and triggers escalation
Role System:
- Users (seek support)
- Volunteers (assist in SOS situations)
- Super Admin (demo monitoring and safety oversight)