How doesNQ work?
The roles of retrieval, machine-learning hints and language-model synthesis.
Give the question context
Include product family, software revision, symptoms and constraints. NQ supports automation, PLC/DCS, protocols, data engineering and industrial AI. An ambiguous question may need clarification.
Source retrieval
The server loads the highest version of each dataset family: 32 active families and 4,213 records in this release. The AI engineering collection adds 304 source-linked sections across two languages, and the question-page collection 104 short-answer records. Turkish normalization and token-level BM25 retrieval consider exact tokens and query coverage.
Local ML, remote synthesis
Browser classifiers and the WebAssembly kernel produce topic and intent hints. These hints are not evidence. Questions go to the NQ server; a remote language model through OpenRouter may synthesize the selected sources. Automatic mode may also retrieve Wikipedia and RSS. Do not include confidential site data, passwords or personal data.
Inspect the answer
Open full source text, document links, dataset versions and review dates when available. Relevance and quality scores are not probabilities of correctness. Legacy collections may lack primary provenance for individual records. If synthesis is unavailable, inspect the retrieved results and reported error.
Validation scope
The expected guide appears in the first three results for 12 Turkish/English retrieval regression questions. This small suite was authored alongside the content; it is not independent field accuracy evidence. Verify critical settings against vendor documentation, site procedures and a qualified engineer.