LLM configuration
The LLM configuration connects an institute to a language model that generates exam questions and supports AI-assisted grading. There is one configuration per institute, and it is the only way to create questions for built-in SES exams.
Provider settings
| Field | Description |
|---|---|
| Provider | openai, azure_openai, mistral, anthropic, google or custom |
| Model | Model identifier (e.g. gpt-4o) or Azure deployment name |
| API key | Encrypted at rest; blank on edit keeps the existing key |
| Base URL | API endpoint (auto-set for standard providers) |
| API version | Required for Azure OpenAI only |
| Max tokens | Maximum response tokens (default 4096) |
| Temperature | 0.00–9.99 (default 0.7) |
Prompt settings
| Prompt | Purpose |
|---|---|
| System prompt | Instructions for question generation |
| Negative prompt | Things the LLM must NOT do |
| Grading system prompt | Instructions for AI-assisted grading |
| Grading negative prompt | Constraints for AI grading |
Starting a generation
- Open the Exam Assembly page of a SES exam
- Click Start AI generation
- Set Question count (5–100), a type mix, and a Bloom's-level distribution (each part must sum to 100%)
- Set a Language (nl, en, de, fr) and an optional Exam AI prompt
- Upload source documents (PDF, DOCX, PPTX or plain text)
- Save — generation runs in the background (watch it under Task status)
Keep the temperature low
For deterministic questions use a low temperature (0.7 or below). Higher values produce more varied but less predictable wording.
Background execution
Generation runs as a background task using Django's native task framework. You can monitor it under Administration → System → Task status.