ClickML.
A drag-and-drop canvas for building ML pipelines without code — rebuilt from a single-file Streamlit app into a multi-service SaaS.
Overview
From a single Streamlit file to a real platform
ClickML shipped in 2023 as one Streamlit script — upload data, pick a model, train, download. That version is still live and is what the public GitHub repo and streamlit.app link reflect today.
The current build is a ground-up rebuild: a proper canvas-based pipeline editor backed by a dedicated execution engine, not a script running top to bottom in a single session.
Fourteen models, wired up as draggable nodes
Pipelines are built visually — data readers, processing steps, feature engineering, and models all connect on a canvas instead of being written as code.
One component map, one execution engine
A PipelineEngine dispatches each canvas node to its implementation through a single COMPONENT_MAP, so adding a new node type is a matter of registering a new component class rather than touching the execution path. Auth is Supabase JWT, forwarded per-request so Postgres row-level security enforces access without a separate authorization layer.
How it works
Four services split the concerns: a canvas frontend for building pipelines, a FastAPI engine for running them, a marketing site, and a docs site — orchestrated together via Docker Compose.