Light

Architecture

Neuroscience & Mental Health - Rootly Sub-Challenge: Engineer Wellbeing

Inspiration

On call engineers operate critical systems under constant interruption, especially during nights and weekends. Every tool in the ecosystem measures system health, but none measure whether the human responding has actually recovered.

Rootly On Call Health captures operational load extremely well. It tells you who is overloaded. But it does not measure whether an engineer has returned to baseline after incidents.

Two engineers can experience the same workload and have completely different outcomes depending on recovery.

We realized the missing layer is not more workload analytics. It is recovery observability.

What it does

Light introduces a new metric:

Recovery Debt

Recovery Debt is the gap between operational load and recovery capacity.

It combines:

Rootly On Call Health score → operational exposure
Recovery signals → sleep and HRV proxies

This produces a single signal that answers:

Are you actually recovering from incidents?

The system provides

Recovery Dashboard

Recovery Debt score with clear states
OCH versus recovery comparison
7 day recovery trend

Incident × Recovery Timeline

Maps incidents to recovery impact
Shows how after hours incidents degrade recovery
Identifies which incident you never recovered from

Circuit Breaker (assisted actions)

Generate handoff summary
Suggest shift override
Escalate to backup

These are recommendations, not automatic actions.

Recovery Mode

Breathing for HRV recovery
Incident debrief journaling
Sleep guidance based on upcoming shifts

How we built it

Mobile app built in Flutter acting as a private recovery sensor
Backend Node.js Express service acting as a Resilience Bridge
ResilienceEngine computes Recovery Debt from operational and recovery signals
Rootly integration using REST API and MCP server for workflow actions
DREAMT dataset used as a proxy for recovery signal modeling

Core logic

Recovery Debt is computed as:

Recovery Debt = Operational Risk − Recovery Capacity

The system models recovery over time, not just static scores.

States:

Stable
Warning
Critical Debt

Challenges we ran into

The biggest challenge was designing a Recovery Debt signal that is simple but meaningful.

A naive linear model does not reflect how recovery behaves. We addressed this by incorporating recovery windows after incidents instead of treating all signals as static.

Another challenge was scope. We intentionally reduced complexity by focusing on recovery rather than full burnout prediction.

Privacy was also critical. We designed a strict boundary: