13A multi-tenant Pharma Manufacturing Analytics SaaS

No-Code AutoML Pipeline Builder

Core contributor

A no-code tool that lets process engineers build ML models on batch data. They draw the pipeline as a graph of imputers, encoders, scalers and estimators. I built the engine that detects variable types and compiles that graph into a valid scikit-learn pipeline from a configurable catalog.

No code: engineers draw the pipeline as a graph
Auto-typed: numeric or categorical, inferred
Always valid: compiles to a scikit-learn pipeline

Python / scikit-learn

Problem

Process engineers on the platform wanted to predict a quality attribute from process parameters without writing scikit-learn code. The UI needed a visual graph of preprocessing and model steps that would always compile into a valid, reproducible pipeline.

Approach

I built the model builder that turns selected batch data and a target into a visual graph of preprocessing and model steps.
Variable types are detected automatically, so each step applies only where it makes sense.
The menu of steps and models is configuration-driven, so new ones need no code.
I built the analysis APIs behind the screens, from target selection to results.

Architecture

Batch data
Target & typing
Visual graph
Model catalog
Pipeline definition
scikit-learn run

Outcome

Engineers build models on batch data without writing code, and the graph they draw compiles to a valid scikit-learn pipeline. The catalog-driven design lets the model menu grow without code changes.

Building something like this?

Tell me about it