AIDflow

Human-in-the-Loop Scorecard Orchestration for Regulated ML

Most automated machine learning tools act as black-box wrappers—they optimize metrics in isolation while hiding the underlying logic, producing models that risk failure during regulatory scrutiny. AIDflow is built differently.

Designed specifically for high-stakes, highly regulated industries (Fintech, Banking, Insurance), AIDflow treats machine learning models not as opaque binary artifacts, but as auditable, legal assets. It provides a comprehensive ecosystem for scorecard development built around Human-in-the-Loop Governance.

Core Pillars

Absolute Micromanagement

Automation should never mean a loss of control. While AIDflow automates repetitive pipeline mechanics, every layer remains fully transparent and configurable. Practitioners can pause execution, inspect intermediate states, and fine-tune parameters—from raw feature ingestion and binning to final champion model selection.

Audit-Ready Reporting

Every execution phase generates comprehensive, standardized documentation out of the box. AIDflow goes beyond traditional accuracy metrics ($AUC$, $Gini$) to produce full validation artifacts required by internal risk committees and external financial regulators.

Production Scorecard Readiness

Outputs extend far beyond standard estimator binaries (.pkl / .joblib). AIDflow generates fully engineered production scorecards complete with Population Stability Index ($PSI$), Characteristic Stability Index ($CSI$), score transformations, and feature explainability vectors that satisfy international credit risk standards (e.g., Basel II/III, IFRS 9).

Bridging the Governance Gap

AIDflow eliminates the friction between rapid experimental prototyping and enterprise compliance—ensuring models that perform in a Jupyter notebook transition seamlessly into compliant production environments.

The Three-Layer Abstraction (The "Super-Gear")

AIDflow decouples the user's intent from the mathematical execution using a three-layer stack. This allows us to change the underlying ML algorithm without ever breaking the User Interface.

Layer 1: The Worker (Interface)

• Role: The Orchestrator.
• Responsibility: Handles UI, input validation, and state registration.
• User Experience: "I want to clean my data" → ctx.workers.fe_cleaner().

Layer 2: The Model (The Gear)

• Role: The Adapter.
• Responsibility: Decouples the Worker from the Fitter. It manages "Fit Options," allowing a single Worker to support multiple different algorithms (e.g., switching between a Sparse GLM and an ElasticNet).

Layer 3: The Fitter (The Engine)

• Role: The Compute Engine.
• Responsibility: Pure mathematical execution. This is where the actual matrix multiplication, optimization, and fitting occur. It has no knowledge of the UI or the Project Context—it only knows data and parameters.

Quick Start

Installation

pip install aidflow

Basic Usage

Defining the Project Context (Control Plane) & Project Scope:

import aidflow as aid

ctx = aid.ProjectContext()

ctx.genesis(
    project      = 'test',
    input_data   = 'test.csv',
    target       = 'target',
    yy_mm        = 'decision_date',
    yy_mm_format = '%d%b%y'
)
ctx.params()

Worker → Model

Accessing worker's inputs schema → parameters & types:

ctx.workers.data_loader.inputs

Accessing worker's fitter options and schema → parameters & types:

ctx.workers.data_loader.fitters

Executing worker with defaults → auto provided by the context:

ctx.workers.data_loader()

Once worker has been executed, thus its main model has been built,
we have access to all populated attributes.

ctx.workers.data_loader.inputs.inputs
ctx.workers.data_loader.result
ctx.workers.data_loader.state
ctx.workers.data_loader.features
ctx.workers.data_loader.stats
ctx.workers.data_loader.params
ctx.workers.data_loader.fitter  # jump to chosen internal fitter
...

View to model report of executed worker:

ctx.workers.data_loader.report.show()

Model → Fitter

Once worker has been executed, we obtain internal fitter state and its parametrization:

ctx.workers.data_loader.fitter

The fitted fitter provides all populated attributes.

ctx.workers.data_loader.fitter.stats
ctx.workers.data_loader.fitter.params

Direct choice and parametrization of internal fitter:

fx  = ctx.workers.data_loader.fitters[0]
prm = dict(
    flow_vars = ['target1', 'target2'],
    columns_subset = ['fe1', 'fe2']
)
loader = ctx.workers.data_loader(fx_choice = fx, fx_params = prm)

In the experimenting phase when various targets might be under testing
(note that target2 must be present in the data):

cfg = dict(
    target = 'target2',
)
woe =  ctx.workers.data_loader(fx_config = cfg)