Forkable references for evaluating real AI applications with AICertify.
A minimal end-to-end demo of the AICertify API.
python examples/quickstart.pyCreates a sample app, adds a few interactions, evaluates against the EU AI Act, and writes a report into reports/. Read quickstart.py before adapting it.
Each folder is a self-contained reference you can copy as the starting point for evaluating your own AI application. The shape is the same in every example so the pattern is easy to follow:
example-name/
├── README.md How to run + how to adapt
├── input_contract.json AI application contract (model + interactions + metadata)
├── sample_interactions.json Standalone interaction set you can splice into a contract
├── policy_config.yaml Which gopal policies + evaluators to run against
├── run.py Runnable script using the Python API
└── expected_report.md What a successful run looks like
| Example | Risk class | Primary frameworks |
|---|---|---|
customer-support-bot/ |
Limited risk | EU AI Act transparency obligations + global baselines |
healthcare-triage-bot/ |
High risk (Annex III) | EU AI Act high-risk + gopal healthcare patient-safety |
hiring-screening-bot/ |
High risk (Annex III) | EU AI Act high-risk + fair-lending proxy + global fairness |
The community is welcome to contribute additional examples following the same shape. Open issues track current asks:
- FastAPI integration example
- LangChain integration example
- LlamaIndex integration example
- Financial-advice bot
- EdTech tutor
- Docker quickstart
See the good first issue and help wanted labels.
If you want to see the AICertify deliverable before installing anything, the outputs/ directory has historical reports from real runs:
outputs/eu_ai_act/— EU AI Act evaluationsoutputs/loan_evaluation/— fair-lending evaluationsoutputs/medical_diagnosis/— patient-safety evaluations
A clean one is also bundled as docs/demo-report-eu-ai-act.pdf.
When you add an example:
- Match the directory layout above. The shape matters more than the content; it's what makes the examples forkable.
- The
metadatablock ininput_contract.jsonmust declare jurisdiction, risk class, and (if Annex III) the relevant subpoint. policy_config.yamlmust include arationale:for each framework explaining why that framework applies.expected_report.mdshould describe both the pass case and the common failure modes a fork might hit.- Be honest about scope. A green AICertify report is necessary but not sufficient for production deployment — say so explicitly.