DF-01Foundational
Field Document OCR Pipeline
Turns decades of scanned reports, logs and maps into searchable text, including Persian and Arabic documents and handwritten annotations.
Data FoundationAI
KC-01FoundationalWave 1AI
Technical Knowledge Assistant (RAG)
In one sentence
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.
The problem
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.
How it works
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.
Data required
- 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
What you get
- 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
What it cannot do
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.
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-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-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 - KC-02Frontier
Domain-Specific Language Model
A model further trained on petroleum engineering text, so it uses the field’s vocabulary correctly in Persian and Arabic as well as English.
Knowledge & CopilotAI