مانا بترو
KC-01أساسيةالموجة 1الذكاء الاصطناعي

مساعد المعرفة الفنية (RAG)

في جملة واحدة

Ask a question about your own field in Persian, Arabic, English or Spanish and get an answer with the report and page it came from.

المشكلة

The answer to most operational questions already exists somewhere in the operator’s own archive. Finding it requires knowing which report to look in, which in practice means asking the one engineer who has been there twenty years.

كيف تعمل

Retrieval-augmented generation over the operator’s indexed archive, running entirely on on-premise models. Retrieval is hybrid — dense and lexical — with domain-aware chunking that keeps tables and figure captions intact. Every answer must cite its sources; an answer the system cannot ground in a retrieved document is refused rather than generated, which is the single most important design decision in this capability.

البيانات المطلوبة

  • Indexed document archive from the OCR and classification pipeline
  • On-premise GPU capacity for embedding and generation
  • Access control model defining which users may see which documents
  • A question set from domain experts for evaluation

ما الذي تحصل عليه

  • Natural-language answers with citations to document, page and passage
  • Source passages displayed alongside every answer for verification
  • Query log showing what people actually ask, which guides archive priorities

ما لا تستطيع فعله

The assistant only knows what is in the indexed archive. It cannot answer a question whose answer was never written down, and it will say so rather than infer. Retrieval quality in Persian and Arabic is measurably below English for the same corpus size, and multi-hop questions that require combining three documents remain a weak point across the industry, not only here.

كل صفحة قدرة تذكر حداً حقيقياً. إن لم يستطع المورّد إخبارك أين يتوقف منتجه، فلم يُقل لك الحق عن أين يبدأ.

قدرات ذات صلة