Open-source ML audit infrastructure for deterministic, audit-ready model reports.
GlassAlpha is an open-source toolkit for deterministic audit reports for tabular ML models, with fairness, explainability, calibration, robustness, and reproducibility metadata in one local workflow.
Operational shape
How the project supports decisions, context, and execution.
GlassAlpha is an open-source toolkit for deterministic, audit-ready reports for tabular ML models.
The workflow is intentionally operational: run a CLI or Python API against a model and dataset, produce reproducible HTML/PDF audit artifacts, and keep the evidence needed to explain model behavior later.
The project focuses on the parts of ML validation that are easy to underbuild: lineage, repeatability, fairness checks, calibration, robustness, preprocessing verification, and deployment gates.
The technical value is in making model validation repeatable enough to inspect, rerun, and gate.
The architecture is deliberately boring: model and data go in, a deterministic audit pipeline runs, and evidence comes out in a form that can support review.
GlassAlpha turns model validation into reproducible audit artifacts, evidence packs, and optional deployment gates.
The project is strongest where reviewers look for execution discipline: deterministic outputs, lineage, gates, and honest limits.
Byte-stable HTML reports on the same platform, Python version, and configuration where supported.
Git SHA, config hash, data hash, seeds, and package versions are captured in the audit trail.
Demographic parity, equal opportunity, predictive parity, and intersectional fairness checks.
Feature importance and SHAP-based individual explanations for supported model classes.
Adversarial perturbation checks and demographic shift simulation.
ECE, Brier score, calibration curves, and confidence-aware calibration analysis.
Fail-on-degradation checks for CI/CD workflows.
Runs offline without an external service dependency.
Public code, docs, architecture notes, examples, and project status are linked directly when they can be shared responsibly.
The repository contains the Python package, CLI, docs, examples, tests, and workflow evidence needed to evaluate the project as software rather than a demo.
These are the practical workflows and constraints the project is built around.
Packages performance, fairness, explainability, calibration, robustness, and reproducibility metadata into one inspectable workflow.
Runs locally so teams can generate evidence without sending model data through an external service.
Supports degradation checks and standardized exit behavior so model quality can become part of deployment review.
GlassAlpha is useful evidence of applied AI systems work because it focuses on reproducibility, audit trails, deterministic outputs, validation gates, and documented model behavior. Those primitives show up wherever oversight and repeatability matter more than demos.
The constraints are explicit, which makes the project easier to evaluate and safer to adapt.
Different surfaces, same focus on context, coordination, review, and execution. This is a selected public view, not a complete project history.
AlignTrue is an open-source reference architecture for supervised AI systems that need durable receipts, replayable state, governed actions, and fenced side effects.
A private MVP build for approved-publication retrieval, voice query, and source-cited maintenance guidance on phone, web, and wearable surfaces.
Main Character helps turn complex product, workflow, and architecture problems into systems that work under real constraints.
