FlowForge is a full-stack, visual workflow-automation platform in the spirit of n8n/Zapier/Make: users drag-and-drop nodes onto a canvas, wire them together with conditional logic, and the backend executes the resulting graph — resolving dynamic expressions between nodes and streaming live status back to the canvas as each node runs.
Designed a custom DAG-based execution engine using Kahn's algorithm that validates and runs workflows of 50+ nodes, executing up to 10 independent branches concurrently and cutting end-to-end run time by 60% versus sequential execution.
Shipped a drag-and-drop canvas with 12+ node types (triggers, conditionals, AI agents, Telegram and Resend integrations), dynamic variable resolution and 3-attempt exponential-backoff retries that reduced failed runs by 40%.
Delivered live execution tracking over Server-Sent Events with under 200 ms update latency, backed by 50+ unit and integration tests.
What It Does
Drag-and-drop canvas (React Flow) for building workflows out of Trigger, Action and Logic nodes, each with its own configurable credentials and parameters.
Execution engine (@workspace/execution-core) traverses the workflow graph, resolves dynamic expressions between nodes (e.g. pulling a previous node's output), and executes each node's logic in sequence.
Live execution feedback over Server-Sent Events — nodes light up and show their output on the canvas as the run progresses, instead of waiting for the whole workflow to finish.
Project-based organization: workflows live inside projects, each with their own credentials scope.
Architecture
A Turborepo monorepo with a single Next.js app (apps/web) hosting the dashboard, canvas and API routes, backed by shared packages: execution-core (the engine that runs workflows and resolves expressions), node-base (base classes/types for standard nodes), validators (Zod schemas for API inputs and node configs), and a shared shadcn/ui component library.