MLAtlas for teams that need more than a global AUC score.
MLAtlas by HickmaTech is a governance-ready command center for post-training audits on classification models and scored datasets. MLAtlas maps score behavior across Segments, Patterns, and Intersections to reveal robustness and reliability risks that global metrics can miss.
Serious model review needs a command surface, not scattered exports.
Audit evidence should come from one live workflow, not scattered scripts and exports.
MLAtlas turns one score-driven audit submission into a tracked organization-scoped run, generated summaries, and downloadable v2 evidence artifacts.
Reviewers need robustness, pattern, and intersection findings in one place.
MLAtlas Segment Diagnostics, MLAtlas Pattern Diagnostics, and MLAtlas Intersection Diagnostics keep global metrics, segment findings, pattern groups, and concentrated intersection signals connected to one audit record.
Teams need careful governance language, especially when higher-risk modes are gated.
MLAtlas maps score behavior across Segments, Patterns, and Intersections to reveal robustness and reliability risks that global metrics can miss.
A short workflow that still produces governance-grade outputs.
Upload a scored dataset
Run score-driven audits with a CSV or CSV.GZ dataset, a binary target, and a numeric score column in [0, 1].
Control the evaluation scope
Choose target, score, optional ID, excluded columns, optional protected columns, and safe v2 audit parameters.
Review live evidence outputs
Track the queue, inspect the generated findings, and download the final artifacts from the completed run.
MLAtlas Score Audit Engine runs two active audit paths with one gated higher-risk mode.
Upload a scored dataset with a score column, map the binary target, and control the v2 audit parameters directly.
Future sandboxed model execution can generate a score and then feed the same v2 audit engine.
MLAtlas Evidence Pack for technical and policy review.
Trustworthy by default, careful about higher-risk workflows.
Built as a paid command center, not a hobby audit tool.
Start with one reviewed pilot workspace, then scale into deeper governance controls.
Early users request access first. After manual approval, HickmaTech enables account creation so the team can sign in, start an audit, review findings, and export the MLAtlas evidence package.
Trial, Starter, Pro, and Custom paths sized by usage and review needs.
Trial is for limited evaluation, Starter is for small teams, Pro is the main validation plan, and larger or regulated needs route to Custom.
Trial
Evaluation-only access for testing the workflow with limited synthetic or local data.
Starter
First paid plan for small ML or data teams running bounded validation workflows.
Pro
Main validation and analytics plan for recurring model-evidence review.
Custom
For regulated teams, larger datasets, higher-volume API users, security requirements, and private/VPC deployment.