Service · Document Q&A
Ask your engineering documents, in Arabic or English
Neuractor builds question answering over your own specifications, procedures, standards and contracts (retrieval-augmented generation, or RAG): a searchable knowledge base built from your own files. Your engineers ask in Arabic or English; the answer comes from your files and shows the document and page, so anyone can check it.
or write to info@neuractor.com
What is the minimum coating thickness for buried pipework?
The example spec asks for a minimum dry film thickness of 400 µm on buried pipework. Spec_EX-201, p. 37, §7.4.2
الحد الأدنى لسماكة الطبقة الجافة 400 ميكرومتر للأنابيب المدفونة. المواصفات، ص 37، البند 7.4.2
Asked next: "Which coating system is approved?" Not in these documents, so it was passed to your coatings lead.
How it works
Four steps, all inside your accounts
- Collect the documents that matter
We agree with your team which specifications, procedures, standards and contracts go in, and which versions are current.
- Read them with page positions
Each document is read page by page, Arabic and English alike, so every passage keeps its document, page and clause.
- Answer only from those documents
A question in Arabic or English is answered from the passages found, with the source shown next to each statement.
- Say "not found" when it isn't there
If the documents don't answer the question, the workflow says so and points to who on your team should answer.
The workflow
What goes in, what comes out, what stays human
In
- Project and company specifications
- Procedures and method statements
- Standards your firm is licensed to use
- Contracts and their annexes
- Manuals and datasheets, including oil and gas procedures (see Neuractor Systems)
Out
- Answers in Arabic or English
- Document, page and clause for each statement
- "Not found" instead of a guess
- A log of questions nobody could answer yet
Stays with your team
- Engineering judgement
- Approvals and sign-offs
- Which documents are current
- Anything that affects safety
What you get
What the pilot delivers
The exact scope is fixed in writing at the end of the 5-day diagnostic.
- Question answering over an agreed set of your documents, running in your own cloud account.
- Answers in Arabic or English, with document, page and clause next to each statement.
- "Not found" handling: questions the documents don't answer are flagged for a named person instead of being filled in.
- A log of unanswered questions, so you can see which documents are missing.
- Reading quality and answer accuracy measured against written acceptance criteria, on your files.
- Handover notes, so your team understands what was built and how to keep the document set current.
Arabic is the hard part
We test your real documents before we promise
Clean, text-based files are the easy case. Scanned Arabic documents, stamped pages and tables are harder: a public benchmark of Arabic OCR and document understanding (KITAB-Bench, ACL 2025) reports that the best-performing model reached only 65% accuracy in PDF-to-Markdown conversion.
So the diagnostic measures reading quality on a sample of your own files, and the pilot's acceptance criteria are written against those files, not against a demo.
Read more: Arabic OCR on scanned tender documents: what breaks and how to test it
FAQ
Questions about document Q&A
Can we ask in Arabic about an English document?
Yes. You can ask in Arabic or English, and the answer cites the page in whichever language the source is written.
Which file types can it read?
Text-based PDFs, Word and Excel files are the usual case. Scanned PDFs work when the scan is readable. The diagnostic tests a sample of your real files before we commit to anything.
What happens when the answer isn't in the documents?
It says so. The workflow marks the item "not found" and passes it to a person on your team instead of guessing.
What if the AI gets something wrong?
Every line points to its source page and clause, so a reviewer can check it against the document. During the pilot we measure accuracy against acceptance criteria agreed in writing during the diagnostic.
Where are our documents stored?
In your own cloud account. We build the pilot inside cloud and AI-service accounts your firm owns, and we tell you in writing which services the workflow uses and where your data goes.
Is our data used to train AI models?
We switch model training off on every work account we use, and your documents stay in your own cloud account.
Neuractor · AI & Systems
The reading is ours. The decision is yours.
Start with one workflow.
Tell us which workflow is slow, manual and full of documents. We reply with whether AI can help and what a diagnostic would look at. No obligation, no mailing list.
info@neuractor.com