Stop Writing the Same Pandas Boilerplate: How We Built a Visual Pipeline Studio for ML Preprocessing

Hariesh Kai — Fri, 09 Oct 2026 02:09:08 +0000

Hey everyone,

Every time I start a new machine learning project or Kaggle competition, I end up spending the first hour doing the exact same chores:

handling null values and missing label rows
writing manual IQR outlier clipping formulas
coercing corrupt object columns to numeric/datetimes
one-hot / label encoding categories
trimming invisible whitespace

I built DataForger to make preprocessing fast, modular, and visual:

What it does:

Drag-and-Drop Ingestion: Instant health check, missing value diagnostics, and distributions.
Visual Node Pipeline: Drag, connect, and reorder transformation stages on a React Flow graph canvas.
Live Diff Previews: Inspect exactly what changed before and after applying an operation.
Export Clean Artifacts: Download the cleaned .csv + the reproducible JSON pipeline configuration to use in your training code.

Early Access & Feedback:
I'm rolling this out as a Founding Member Early Access Pass for ₹299 INR (~$3.50 USD one-time lifetime) for the first 50 users before we switch to a subscription model.

Try it out here: [https://shirogani-dataforger-beta.vercel.app]

I really want your raw feedback:

What's the most annoying preprocessing step in your daily workflow that you wish was automated?
Does the node graph feel snappy or would you prefer a linear list view?

I'll be in the comments all day to answer questions and patch any bugs you find!

DEV Community: Hariesh Kai

Stop Writing the Same Pandas Boilerplate: How We Built a Visual Pipeline Studio for ML Preprocessing