DF-02Foundational
Structured Extraction from Reports
Pulls the numbers out of the reports — depths, rates, pressures, volumes, dates — and puts them into a database with a link back to the page they came from.
Data FoundationAIDomain
DF-01FoundationalWave 1AI
Field Document OCR Pipeline
In one sentence
Turns decades of scanned reports, logs and maps into searchable text, including Persian and Arabic documents and handwritten annotations.
The problem
A typical operator holds tens of terabytes of scanned documents: daily reports, end-of-well reports, laboratory studies, correspondence, vendor reports. None of it is searchable, so the practical answer to any historical question is "we would have to go and look", which means the question is not asked.
How it works
A staged pipeline handles deskewing, denoising and layout analysis before recognition, with separate models for printed Latin, printed Persian and Arabic script, tables and handwriting. Persian and Arabic recognition is the hard part and is handled by models tuned on documents from this industry rather than by a general-purpose engine. Every page carries a per-region confidence score, and low-confidence pages are routed to human verification instead of being silently accepted.
Data required
- The document archive in whatever state it exists, including mixed-quality scans and photographs of paper
- Storage and compute on premise, sized to the archive
- A sample set of representative documents for tuning, ideally several hundred pages
- A naming or foldering convention if one exists, to seed metadata
What you get
- Searchable text layer written back into the PDFs, with the originals untouched
- Per-page confidence scores and a verification queue for low-confidence pages
- Extraction-ready output feeding structured extraction and the retrieval index
What it cannot do
Recognition accuracy on clean typed Persian text is high; on handwritten Persian in a remarks column it is much lower, and those regions are flagged rather than trusted. Blueprint-style drawings and well-log images are indexed as images with their text captions — the curves themselves are not digitised by this capability.
Every capability page states a real limitation. If a vendor cannot tell you where their product stops, you are not being told the truth about where it starts.
Related capabilities
- DF-03Foundational
Archive Auto-Tagging & Classification
Works out what each document is and which well, field and date it belongs to, so the archive becomes navigable rather than a folder tree.
Data FoundationAI - DR-13Foundational
NPT Root-Cause & Offset Well Lookup
Reads every daily drilling report the operator has ever filed and answers, in seconds, what went wrong at this depth in this field before.
Drilling IntelligenceAI - HS-05Foundational
Equipment Nameplate & Tag OCR
A technician photographs a nameplate and the equipment record is created or matched, instead of being typed from memory later.
HSE, Emissions & SecurityAI - KC-01Foundational
Technical Knowledge Assistant (RAG)
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.
Knowledge & CopilotAI