Core Concepts

Architecture Overview

Design principles and framework layout of Featuresmith

Featuresmith is engineered around a core-first layout. All critical logic is centralized, ensuring surfaces remain thin rendering components.

Architectural Design Rules

One core, many thin surfaces: All business logic — loading, profiling, rule validation, custom rules — resides inside the package featuresmith-core. Interfaces (CLI and Dashboard) only wrap the SDK api surface.
Compute and reasoning separation: Numerical algorithms and database scans run deterministically using vectorized backends. AI integrations (planned for Phase 2+) are only ever used for natural-language narration or chat - they never perform computation.
Modular plugin structure: Connectors, rules, exporters, and AI providers inherit from stable base classes (e.g. BaseRule, BaseConnector), allowing third-party extensions without altering repository core code.

Data Pipeline Blueprint

The flow of execution runs in a linear sequence, transitioning between structured stages:

bash

1[Raw Dataset / DataFrame]
2           │
3           ▼  (Connectors Load)
4    [Dataset Object]
5           │
6           ▼  (Profiling Engine Summarizes)
7  [ProfileResult Dataclass]
8           │
9           ▼  (Rule Engine Audits)
10  [RuleResult (Profile + Findings)]
11           │
12           ▼  (Exporter / Chat Integration)
13[Sklearn Pipelines / Text Reports / FAQ response]