kwikly_py

This repo trains and scores a logistic regression model that predicts whether a dental temporary staffing shift will be filled. It ingests a 68-column CSV training dataset exported from a BI pipeline, fits a baseline classifier, and emits metrics and artifacts (model weights, predictions). The codebase is purpose-built for the Kwikly platform—a W-2 dental staffing marketplace connecting dental offices with hygienists, assistants, and other professionals—and is designed to support operational research documented in research.md rather than production deployment.

Built to establish a machine-learning baseline for shift fill prediction as part of an agentic software engineering learning exercise. The project explores how well simple features (shift timing, location, pay, contractor availability) can forecast marketplace liquidity, with the goal of informing product decisions (pricing, notification timing, supply/demand balancing) and testing ML-ops patterns (schema validation, train/score CLIs, time-based holdouts).

Python 3.11+, pandas (data wrangling), scikit-learn (LogisticRegression), joblib (model serialization), pytest (testing). Packaged as an editable install with pyproject.toml; includes CLI entry points (shift-fill-train, shift-fill-score) and small non-sensitive test fixtures.

Last active May 21, 2026 — 6 commits across 3 active days (spanning 2 days).

Quick research spike over 3 days in mid-May 2026 (6 commits). This is a focused prototype for a single ML experiment—not an ongoing project.

The repo delivers a working logistic baseline with train/score CLIs, tests, and a schema that aligns with the upstream BI export. However, it explicitly exists for research (research.md) rather than production use, the model artifact is not committed, and there is no CI or deployment config. The 6-commit span and "research workspace" framing signal a targeted investigation rather than a shipped tool.