Skip to content

Dokimi

questions

`dokimi questions`¶

usage: dokimi questions [-h] {generate,analyze,refine,panel-review} ...

positional arguments:
  {generate,analyze,refine,panel-review}
    generate            Generate questions from personas and topics
    analyze             Analyze question quality (difficulty + triviality)
    refine              Refine question set: normalise, deduplicate, filter trivial, report gaps
    panel-review        Multi-agent panel review of question quality

options:
  -h, --help            show this help message and exit

`dokimi questions generate`¶

usage: dokimi questions generate [-h] --product PRODUCT --personas PERSONAS [--output OUTPUT]
                                 [--topics TOPICS] [--top-k TOP_K] [--count COUNT] [--validate]
                                 [--embeddings {auto,openai,local}] [--model MODEL]
                                 [--provider {openai,anthropic,amazon-bedrock,google-vertex,openrouter,openai-codex}]
                                 [--doc-source DOC_SOURCE] [--context7-id CONTEXT7_ID]

options:
  -h, --help            show this help message and exit
  --product PRODUCT     Product name (e.g., oneTBB)
  --personas PERSONAS   Path to personas JSON file
  --output OUTPUT       Output file (default: questions/{product}.json)
  --topics TOPICS       Optional: path to topics JSON (auto-extracted if not provided)
  --top-k TOP_K         Docs to retrieve per topic (default from config)
  --count COUNT         Questions per topic per persona
  --validate            Enable validation and deduplication
  --embeddings {auto,openai,local}
                        Embedding backend for deduplication: openai, local (sentence-transformers,
                        no API key), or auto (openai when a key is set, else local)
  --model MODEL         LLM model for generation
  --provider {openai,anthropic,amazon-bedrock,google-vertex,openrouter,openai-codex}
  --doc-source DOC_SOURCE
                        Documentation source for topic extraction: 'context7', 'local:<path>',
                        'url:<url>'
  --context7-id CONTEXT7_ID
                        Explicit Context7 library ID. Overrides auto-resolution.

`dokimi questions analyze`¶

usage: dokimi questions analyze [-h] --questions QUESTIONS --product PRODUCT [--output OUTPUT]
                                [--model MODEL]
                                [--provider {openai,anthropic,amazon-bedrock,google-vertex,openrouter,openai-codex}]
                                [--concurrency CONCURRENCY]

options:
  -h, --help            show this help message and exit
  --questions QUESTIONS
                        Path to questions JSON file
  --product PRODUCT     Product name (e.g., oneTBB)
  --output OUTPUT       Output report JSON (default: reports/{product}_question_quality.json)
  --model MODEL         LLM model for classification
  --provider {openai,anthropic,amazon-bedrock,google-vertex,openrouter,openai-codex}
  --concurrency CONCURRENCY
                        Parallel classification requests

`dokimi questions refine`¶

usage: dokimi questions refine [-h] --questions QUESTIONS --product PRODUCT [--output OUTPUT]
                               [--target-beginner TARGET_BEGINNER]
                               [--target-intermediate TARGET_INTERMEDIATE]
                               [--target-advanced TARGET_ADVANCED] [--sim-threshold SIM_THRESHOLD]
                               [--dry-run] [--fill-gaps] [--fill-model FILL_MODEL]
                               [--fill-provider {openai,anthropic,google}]

options:
  -h, --help            show this help message and exit
  --questions QUESTIONS
                        Path to questions JSON file
  --product PRODUCT     Product name (e.g., oneTBB)
  --output OUTPUT       Output refined JSON (default: questions/{product}_refined.json)
  --target-beginner TARGET_BEGINNER
  --target-intermediate TARGET_INTERMEDIATE
  --target-advanced TARGET_ADVANCED
  --sim-threshold SIM_THRESHOLD
                        Edit-distance similarity threshold for deduplication (default: 0.82, range
                        0-1)
  --dry-run             Print report only, do not write output file
  --fill-gaps           Use LLM to generate questions for under-represented difficulty levels
  --fill-model FILL_MODEL
  --fill-provider {openai,anthropic,google}

`dokimi questions panel-review`¶

usage: dokimi questions panel-review [-h] --questions QUESTIONS --product PRODUCT
                                     [--output OUTPUT] [--model MODEL]
                                     [--provider {openai,anthropic,amazon-bedrock,google-vertex,openrouter,openai-codex}]
                                     [--reviewers REVIEWERS] [--concurrency CONCURRENCY]
                                     [--limit LIMIT]

options:
  -h, --help            show this help message and exit
  --questions QUESTIONS
                        Path to questions JSON file
  --product PRODUCT     Product name (e.g., oneDAL)
  --output OUTPUT       Output report JSON (default: reports/{product}_question_panel.json)
  --model MODEL
  --provider {openai,anthropic,amazon-bedrock,google-vertex,openrouter,openai-codex}
  --reviewers REVIEWERS
                        Comma-separated reviewers (default:
                        domain_expert,user_advocate,qa_engineer)
  --concurrency CONCURRENCY
  --limit LIMIT         Review only first N questions