No ill-formed inputs
No SSN
No credit card information
No ill-formed inputs
No SSN
No credit card information
No ill-formed inputs
No SSN
No credit card information
Is valid JSON
Is valid Python code
No column drift
Is valid JSON
Is valid Python code
No column drift
Is valid JSON
Is valid Python code
No column drift
High GPT-evaluation score
High answer relevancy
Response is concise
High GPT-evaluation score
High answer relevancy
Response is concise
High GPT-evaluation score
High answer relevancy
Response is concise
Response is factual
High BLEU-1 score on protected subpopulation
High METEOR score
Response is factual
High BLEU-1 score on protected subpopulation
High METEOR score
Response is factual
High BLEU-1 score on protected subpopulation
High METEOR score
No ill-formed sentences
No new tokens
No disparity between gender-related pronouns
No ill-formed sentences
No new tokens
No disparity between gender-related pronouns
No ill-formed sentences
No new tokens
No disparity between gender-related pronouns
No rows from the training set present in validation
No new labels in the validation set
No label drift
No rows from the training set present in validation
No new labels in the validation set
No label drift
No rows from the training set present in validation
No new labels in the validation set
No label drift
No significant label drift
No rows from the training set present in validation
Expect high performance on sentences containing "help"
No significant label drift
No rows from the training set present in validation
Expect high performance on sentences containing "help"
No significant label drift
No rows from the training set present in validation
Expect high performance on sentences containing "help"
High precision on sentences containing key tokens
High confidence and accuracy on must-pass cases
High precision on "urgent" predictions
High precision on sentences containing key tokens
High confidence and accuracy on must-pass cases
High precision on "urgent" predictions
High precision on sentences containing key tokens
High confidence and accuracy on must-pass cases
High precision on "urgent" predictions
No more than 10 rows with nulls
No more than 10 duplicate rows
No significant difference in accuracy between gender feature values
No more than 10 rows with nulls
No more than 10 duplicate rows
No significant difference in accuracy between gender feature values
No more than 10 rows with nulls
No more than 10 duplicate rows
No significant difference in accuracy between gender feature values
No rows from the training set present in validation
No new labels in the validation set
No null columns
No rows from the training set present in validation
No new labels in the validation set
No null columns
No rows from the training set present in validation
No new labels in the validation set
No null columns
No significant label drift
No rows from the training set present in validation
Expect high performance on young adult females in South Africa
No significant label drift
No rows from the training set present in validation
Expect high performance on young adult females in South Africa
No significant label drift
No rows from the training set present in validation
Expect high performance on young adult females in South Africa
High precision on "fraudulent" predictions
High confidence and accuracy on must-pass cases
High precision on "urgent" predictions
High precision on "fraudulent" predictions
High confidence and accuracy on must-pass cases
High precision on "urgent" predictions
High precision on "fraudulent" predictions
High confidence and accuracy on must-pass cases
High precision on "urgent" predictions
Support for your task type
We support a diverse range of task types so that all of your use cases are covered.
Tests
Make sure your AI is at peak performance when it gets in the hands of your users. Tests are a great way to track all of the constraints that are important for your model's performance.
Data integrity
Start by improving the foundation of your AI system: the data.
Data consistency
Make sure data stays consistent between different datasets.
Performance
Identify underperforming subpopulations Choose from the most advanced metrics.
Fairness
Guard against biases and ensure equal treatment of sensitive groups.
Robustness
Probe for edge-cases not captured by your data that you may encounter in the wild. See how your AI performs under adversarial attack.
Monitoring
Add your monitoring pipeline to set production-specific tests and keep a close eye on your model behavior in the wild.
Real-time alerts
Something went wrong in production? Be the first to know with real-time pings on email, Slack, or in-app.
Openlayer
@sophia commented on No duplicate rows
– thoughts on changing the threshold?
Openlayer
Test status updated for No output drift
To 🔴 Failing From 🟢 Passing
Evaluation windows
Different tests require different windows of data. Set custom evaluation windows to determine when to run each test.
Monitor dashboard
The monitoring dashboard offers a comprehensive view of current test results. Click on any individual test to dive deeper into its performance history.
Diagnosis
Breeze through the "why?"s behind failed tests. The information you need to find the root cause of issues is at your fingertips.
Root cause analysis
Dive deeper into every test to understand why it is failing. Stop questioning what to do next to improve your model.
Explainability
Understand which features are the culprits for driving model performance over a particular data slice or r the whole dataset.
What-if-analysis
Perturb individual model inputs and see how the prediction changes. Compare the explainability scores and model predictions side-by-side to validate your hypotheses.
Versioning & experiment tracking
Track and version your models, prompts, and datasets. Compare performance across versions, and systematically choose the best AI stack.
Prompt playground
Experiment with different models, prompts, and parameters and generate test cases.
Version commits
Keep track of and easily switch between model and dataset versions.
Side-by-side comparison
Quickly compare versions to pick a winner or revert ones that introduce regressions.
Collaboration
Bring the whole team in on the development of your AI. Work with others to diagnosis issues and identify next steps. Keep everyone in the loop.
One shared workspace
Increase visibility with everyone in one place. Everyone stays up to date on the versions and progress towards deployment.
Lee
Jordan
Fatima
Pranav
Yasmin
Josh
Casey
Bianca
Zachary
Charlie
Wesley
Priyam
Aisha
Nathan
Nina
Maya
Andre
Helena
Jared
Rachel
Malcolm
Katie
Maddy
Lila
Jocelyn
Lauren
LucienDiscussions
Work together to create tests and improve your model by leaving comments. Add context by tagging data, features, and more.
How to get started
Sign up for free
Get your free account up and running in 60 seconds
Get a product walkthrough
Our team will give you a demo of our platform and answer any questions
Personalized onboarding
We'll get your and your team's accounts set up according to your specific needs