Salesforce / General-purpose automation platform
EdgeFlow
Make operational knowledge reusable. Give engineers a common platform to build on.
The problem
Engineering scripts were scattered across repositories and run from developers' computers. The team needed shared execution and a way to build operations once, then compose them into many workflows.
What I built and led
I conceived the platform, pitched it to leadership and secured funding. I built the core Python framework and led collaborative delivery from concept through production. Engineers contributed service clients and reusable operations to a growing shared catalog, with workflows scheduled through existing Salesforce-managed Apache Airflow.
The resulting platform supported several dozen use cases and cross-team adoption. It was still operating in production as of my October 2026 account.
How the platform earned adoption
Git and PagerDuty integrations needed an approved infrastructure environment, but the Edge team lacked capacity for Airflow's database and virtual machines. I pitched incident-response automation to a partner team responsible for automation and monitoring, and secured approval for a Salesforce-managed Airflow instance.
The implementation took longer than I had expected. I launched alert automation as Salesforce Edge's growth drove a sharp increase in alert volume. Its results helped convince my team to adopt the platform. Release automation, the use case that had motivated the original design, followed on the same foundation.
One use case: incident response
I built automated triage and recovery, including AI agents that selected remediation actions and created pull requests for human approval where required. PagerDuty owned the independent escalation timer, keeping human fallback available when EdgeFlow was unavailable.
82% of high-severity and 88% of low-severity alerts resolved automatically.
Measured during one month after launch, for PagerDuty alerts routed to automation. These are separate severity populations, not an overall rate. The results cover the entire routed automation population, not AI-only performance.Another use case: AI-assisted guarded releases
I built AI-assisted release orchestration on the same platform so any team member could own a safer release process for Salesforce Edge's tier-0 service. Read how the release system worked.
A production lesson
A first-day alert storm exposed a backlog in one-workflow-per-alert execution. I wrote and deployed a dedicated batch workflow that grouped active alerts of the same type into one PagerDuty storm incident, addressing the immediate processing bottleneck.
Ownership
I originated the platform, built its Python core and led collaborative delivery. Other engineers extended the shared client and operation catalog; Salesforce-managed Airflow provided workflow scheduling.