YC W23
AI Workflow Execution at Stack AI
Simple to compose.Built to execute.
Working underneath the canvas of a visual AI builder: execution, integrations, and the events that connect a workflow to the real world.
Behind the experience
Explore the flowStart with a clear connection.
A visual graph describes the work to be done.
I worked on the abstraction layers that translate graph-based inputs into executable AI pipelines.
Give each node a contract.
Inputs become outputs through a consistent execution model.
I migrated the node engine from an edge-based data-flow model to a cleaner input-output structure, improving maintainability and extensibility.
Make the workflow useful outside the canvas.
Email inputs, outputs, and events connect the system to everyday work.
I built Gmail and Outlook nodes, email-trigger automations, and an in-house notification system for workflow events.
An illustrated overview of the work. Select a stage to look closer.
Compose
01 / The challenge
What had to
feel simple.
A visual workflow makes a complex system look approachable. Behind that canvas, each node needs to accept inputs, execute its work, and pass a useful result along. The simplicity people see depends on the structure they don't.
02 / My contribution
Where I got
my hands dirty.
Reworking the execution model
Migrated the node engine toward a clearer input-output structure, with scalability, maintainability, and new node types in mind.
Bringing email into the graph
Built Gmail and Outlook input/output nodes and email-trigger automations, extending what people could do from the visual builder.
Making events visible
Shipped an in-house notification system that surfaces real-time workflow events to users.
03 / The engineering idea
A clear contract makes room for more possibilities.
When nodes share a consistent input-output structure, the execution model is easier to reason about and extend. The visual surface can stay simple while the range of things it can connect keeps growing.
What stays with me
A good abstraction doesn't hide complexity by ignoring it. It gives that complexity a structure people can trust and build on.
SimCare AI
Real-time AI · Full-stack engineering