Dataxiom

Dataxiom turns datasets into reproducible, interactive data science studies so readers can examine the question, evidence, model choices, results, and limitations together. Formerly listed in this portfolio as Data Science Showcase Platform, it is Axiomizer's applied data-science companion, maintained as a separate Django application.

Problem

Notebooks and static reports often separate model results from dataset provenance, evaluation assumptions, and explanations. Readers need a consistent way to assess the evidence and explore a model's behavior.

Solution

Dataxiom combines structured study publishing, dataset and model registries, offline training, interactive inference, accessible charts, and learning exercises. Its publishing validation checks for question and limitations sections, a reference, a linked dataset, and alt text when hero media is present.

The flagship study uses the UCI Heart Disease Cleveland dataset: 303 records, split into 227 training and 76 test rows. Its served-model design uses 11 features, excluding ca and thal because they are follow-up measurements unavailable in the intended screening scenario.

The source model card reports calibrated logistic regression recall of 0.857, with a 95% bootstrap interval of [0.733, 0.968], precision of 0.857, and ROC AUC of 0.925 on that test split. Logistic regression was chosen for interpretability and small-sample robustness; the study does not establish superiority over the other candidates. These are documented experimental results, not production benchmarks or evidence of clinical validity.

User Experience

Visitors browse and search studies, inspect dataset dictionaries and model cards, explore prediction forms, read charts with accompanying data tables, and take quizzes with answer explanations. Accounts add saved favorites, recorded quiz attempts, and moderated comments. Study reading and charts use server-rendered HTML and SVG.

The heart disease study is an educational demonstration. Its referral-population results do not establish individual or general-population health risk.

Current Status

Deployed at https://dataxiom.sinabolouri.com. On 2026-09-17, the homepage, health endpoint, flagship study, and model card returned HTTP 200. The reviewed code implements publishing, inference, learning, community features, charts, search, and first-party analytics. Production prediction execution was not independently verified during this review.

A project-specific AI assistant, semantic retrieval, shared Axiomizer identity, and background workers remain unimplemented. See STATUS.md for operational gaps.

Tech Stack

Python, Django 5.1, PostgreSQL, scikit-learn, pandas, Tailwind CSS, Gunicorn, WhiteNoise, and Docker. The deployment configuration targets Render with Neon PostgreSQL. JavaScript enhances theme selection and section navigation; HTMX integration remains incomplete.

Evidence: Dataxiom's application code, deployment files, and ml/studies/heart_disease_risk/{MODEL_CARD,DATASET,LIMITATIONS}.md.