Action AI
- Domain
- Enterprise AI automation
- Role
- Front end
A low-code platform where teams build multi-agent AI workflows by connecting nodes on a canvas, then test and run them on real data.
- Built the graph-based workflow editor, where each node carries its own configuration and typed input and output ports, and incompatible connections are blocked while dragging.
- Built real-time collaboration on the canvas with Yjs CRDTs, syncing edits, live cursors and node selections between users.
- Designed the Zustand state layer, with the store split into typed slices and Immer patches powering undo and redo.
- Built the reliability rules UI with regex classifiers and routing, and the dataset mapping grid for running evaluations.
- Built agent nodes including the LLM agent with structured outputs, OCR, deduplication and iterators, each with its own typed configuration panel.
- Added resumable tus uploads with Uppy for large dataset files, and live updates over SignalR for AI-generated workflow descriptions.
- React
- TypeScript
- React Flow
- Zustand
- Yjs
- ShadCN
- TanStack Query
- SignalR