Quantology
Now live
Model Maker
An open-source, visual, AI-assisted pipeline builder for rapid model prototyping — without turning it into a black box.
Model Maker is a visual pipeline builder for financial model development in Python. Blocks of real, functional Python code are wired together on a canvas — organized into phase-labelled swimlanes — and compile down to a single, human-readable, self-contained script.
An embedded, pluggable AI provider can draft and fix blocks for you, so the parts of model development that are mostly boilerplate move faster, while every line of generated code stays visible, editable, and yours.
AI that connects blocks, not AI that writes code
Most AI tools generate new code every time — code someone then has to read, test and debug before it can be trusted. In Model Maker that code already exists and is tested. The AI only decides which blocks to use, how to configure them and how to wire them together, which changes the picture completely.
Faster
With no code to write, a complete PD pipeline, from raw data to Gini, KS and PSI, is planned, built and run in minutes.
More reliable
Every block is pre-tested code, so the AI can't trip over library quirks such as Polars syntax or changing APIs. It runs each block as it goes and fixes failures before moving on.
Safer
The AI never executes code of its own; only the fixed code of the library blocks runs. Guardrails such as excluding post-default columns are enforced by the tools, not just the prompt, and a single Undo reverts the whole build.
No data shared
The AI only ever sees column names, types, roles and aggregate statistics. Individual records never leave your machine.
Built for model risk, not just model building
- Nothing hidden — every pipeline compiles to a single, readable Python script, so what runs in validation is exactly what you can read.
- Built to iterate fast — sample mode caps every source at a handful of rows while you prototype, so a full pipeline stays quick to build and rerun before you point it at the full population.
- Runs on your own machine — your data and models never leave your infrastructure.
- AI-assisted, not AI-locked — draft and fix blocks with Claude, OpenAI, Gemini, a local model, or none at all.
- Both a web UI and a terminal UI, talking to the same backend, so it fits how your team already works.
Why it matters for model risk
Model development tools that hide their logic behind a black box are hard to validate and harder to trust. Model Maker keeps every block's code visible and the compiled output human-readable, so acceleration doesn't come at the cost of auditability.
See it in action
A short walkthrough: loading the demo pipeline, adding a Predict and an AUC / Gini block on the held-out test set, and comparing train vs. out-of-sample Gini in the inspector.
Or build it by hand
A short walkthrough: loading the demo pipeline, adding a Predict and an AUC / Gini block on the held-out test set, and comparing train vs. out-of-sample Gini in the inspector.
Try it
Model Maker is open source and installable via pip.