Ainanza AI Competency Framework: AAI-1 to AAI-3
Three levels, each measuring a different stage of practical AI competence. This page explains exactly what each level proves, how scoring works, and what a badge does and does not mean.
AAI-1: AI Foundations
Core AI knowledge, basic prompting, everyday use, verification, privacy, and responsible AI use.
| Questions per attempt | 20 |
| Estimated time | ~20 minutes |
| Passing score | 70% |
| Minimum safety score | 60% |
| Question bank minimum | 60 reviewed questions |
| Badge prerequisites | None |
| Assessment version | 1.0 |
Someone who passes AAI-1 should be able to:
- Explain basic AI and generative AI concepts
- Recognize important AI terminology
- Write a structured basic prompt
- Identify useful everyday AI tasks
- Recognize hallucinations and weak outputs
- Avoid sharing private or confidential information carelessly
- Verify important information
- Understand the need for human judgment
- Distinguish safe and unsafe everyday AI use
Domains tested: AI Foundations, Prompting, AI Workflows, Research and Verification, Safety and Responsible AI.
Take AAI-1AAI-2: Applied AI Practitioner
Practical prompting, workflows, research, tool selection, output evaluation, and safe professional AI use.
| Questions per attempt | 25 |
| Estimated time | ~30 minutes |
| Passing score | 75% |
| Minimum safety score | 65% |
| Minimum score per competency | 50% |
| Question bank minimum | 90 reviewed questions |
| Badge prerequisites | AAI-1 passed first |
| Assessment version | 1.0 |
Someone who passes AAI-2 should be able to:
- Design structured prompts for specific outcomes
- Improve prompts iteratively
- Turn repeated tasks into workflows
- Select appropriate tools for a task
- Evaluate AI outputs using clear criteria
- Research topics while checking sources
- Identify accuracy, privacy, and bias risks
- Add human-review checkpoints
- Use AI in realistic professional scenarios
- Identify tasks that should not be delegated fully to AI
Domains tested: Prompting, AI Workflows, Research and Verification, Safety and Responsible AI.
Take AAI-2AAI-3: AI Systems Specialist
Advanced knowledge of LLM applications, RAG, retrieval, agents, evaluation, monitoring, architecture, and safety.
| Questions per attempt | 30 |
| Estimated time | ~45 minutes |
| Passing score | 80% |
| Minimum safety score | 70% |
| Minimum score per competency | 60% |
| Question bank minimum | 120 reviewed questions |
| Badge prerequisites | AAI-1 and AAI-2 passed first |
| Assessment version | 1.0 |
Someone who passes AAI-3 should be able to:
- Explain the structure of LLM-powered applications
- Understand embeddings and vector retrieval
- Reason about chunking, retrieval, and reranking
- Design a basic RAG system
- Design agent workflows and tool boundaries
- Identify common agent failure modes
- Create evaluation plans
- Reason about quality, cost, and latency trade-offs
- Define guardrails and approval gates
- Understand production monitoring risks
- Apply responsible AI and governance principles
Domains tested: AI Systems, AI Agents, Evaluation and Monitoring, Safety and Responsible AI.
Take AAI-3Question types
Every level mixes several question formats so the assessment measures more than memorized definitions.
| Type | What it tests |
|---|---|
| Single choice | One correct answer among up to five options. |
| Multiple choice | Several answers can be correct. Full credit only when the complete correct set is selected and nothing else. |
| Fill in the blank | A short, unambiguous term typed from memory. |
| Matching | Pairing concepts with their correct definitions using select inputs. |
| Ordering | Placing steps of a process in the correct sequence. |
| Scenario | A realistic situation with one defensible best answer. |
| Best prompt | Comparing several prompts for the same task and picking the strongest one. |
| Identify the problem | Spotting a specific flaw in an AI output, process, or system. |
| Output evaluation | Comparing two or more AI outputs against stated criteria. |
| Risk classification | Classifying a scenario into one of five risk categories. |
Risk classification categories
Risk classification questions (used mainly in AAI-3, and in scenario questions elsewhere) ask you to sort a situation into one of five categories:
| Category | Definition |
|---|---|
| Low risk | The task has no meaningful downside if the AI output is imperfect. Mistakes are easy to spot and cheap to fix, and nothing private, financial, medical, legal, or safety-related is involved. |
| Moderate risk | A wrong or biased output could cause real but limited harm, for example wasted time, minor reputational friction, or a small factual error reaching a small audience. A quick human check before use is enough. |
| High risk | A wrong output could cause meaningful harm: financial loss, reputational damage, legal exposure, unsafe advice, or discrimination against real people. This work needs careful review before anything is used or acted on. |
| Requires human approval | The action has consequences that should never be taken automatically, for example sending money, publishing publicly, contacting a customer, or changing production systems. A qualified human must explicitly approve before the action happens. |
| Not suitable for AI use | The task requires judgment, accountability, or expertise that current AI tools cannot responsibly provide, such as final medical diagnoses, legal rulings, or decisions that must legally be made by a licensed human professional. |
How passing works
Passing requires more than a high overall score. Each level also has a minimum score specifically for Safety and Responsible AI questions, and AAI-2 and AAI-3 additionally require every tested competency to clear its own minimum. A strong overall score can still result in a "not passed" outcome if one of these minimums is missed, and your result page explains exactly which requirement was not met.
Assessment versioning
Each level carries its own version number, currently 1.0 for all three. Small corrections, explanation updates, or minor question changes bump the version to 1.1 and do not affect saved results. A major framework, scoring, domain, or question-bank change bumps the version to 2.0. If you have an assessment in progress when a major version ships, that attempt is cleared and you are offered a fresh attempt under the new version rather than having old and new question data mixed together.
Where your data lives
Assessment attempts, results, and badges are stored only in your browser (localStorage), not on an Ainanza server. Clearing your browser data, switching devices, or using private browsing will remove your saved progress. There is no account system and no way for Ainanza to recover lost results.
What a badge means
Ainanza badges represent completion of an educational knowledge assessment. They are not an accredited professional certification.