An Interactive 5-Day Training Course
Generative AI for Subsurface Workflows
Generative AI for Subsurface Workflows
Secure your spot now!
Duration
5 Days
Price
$ 3,500
Starting From
26 Jul 2027
Home Geology and Geoscience Generative AI for Subsurface Workflows
Generative AI for Subsurface Workflows
Have Questions ?
Generative AI for Subsurface Workflows - GG-GENAI-PEA27
| Code | Date | Time | Duration | Location | Currency | Early Bird Fee Per Person |
|---|---|---|---|---|---|---|
| GG-GENAI-PEA27 | 26 - 30 Jul 2027 | 10 AM CST | 4 Hours Per Day |
Online |
USD |
3500 |
Need this for a group? We deliver the same course in-house — face-to-face at your location or online — tailored to your assets and team level. Contact info@peassociations.com
Boost your team's skills and your budget! Enjoy group discounts for collaborative learning. Send an inquiry to info@peassociations.com.
Generative AI for Subsurface Workflows
This training covers generative artificial intelligence applied to subsurface workflows. It works through what these models are and how they behave, prompting and task framing, document and report analysis, data extraction from unstructured sources, code generation for technical tasks, retrieval from internal document sets, verification requirements, confidentiality and data handling, limitations and failure modes and the governance of use in technical work.
Description
Generative models are now in use across subsurface organisations for summarising reports, extracting data from documents, drafting text, generating code and answering questions about internal material. They are genuinely useful for several of these tasks and unreliable for others, and the difference matters because subsurface decisions rest on technical accuracy. Using them well requires understanding what they are doing, which tasks suit them and what verification each application needs.
This training covers that practical judgement. What these models are and how they behave is established first, covering their training, their mode of operation and the origin of their characteristic failures, without unnecessary technical depth. Prompting and task framing follow, covering the construction of requests that produce useful output. Document and report analysis is then developed as one of the strongest applications. Data extraction from unstructured sources follows. Code generation for technical tasks is then addressed. Retrieval from internal document sets follows, covering the approach that grounds answers in an organisation's own material. Verification requirements are then developed, task by task. Confidentiality and data handling follow. Limitations and failure modes are addressed. The training closes with governance.
These models produce fluent text whether or not it is correct, and that is the central practical fact. A request for a summary of a report produces a summary; a request for a figure that is not in the report produces a plausible figure that is invented. The output carries no signal distinguishing the two, because the model is generating likely continuations rather than retrieving facts. Every use therefore needs a verification step matched to the consequence of being wrong, and applications where verification is impractical are applications to avoid.
Grounding the model in your own documents changes the reliability substantially. Rather than asking a model to answer from its training, a retrieval approach searches the organisation's own reports and passes the relevant passages to the model with the question, so that the answer is drawn from supplied material and can cite its source. This reduces invention considerably, allows the answer to be checked against the cited passage, and works on internal material the model has never seen. It is the configuration most worth building for technical use.
Code generation is one of the strongest applications because the output is testable. A request for a script to read a directory of log files and produce a summary table produces code that either runs and gives the right answer or does not, and that can be checked in minutes. The failure mode is not invention but subtle error: code that runs and produces plausible but wrong output. Testing generated code against a case with a known answer, rather than reading it and judging it correct, is the verification that matters.
Finally, confidentiality is a practical constraint that shapes what can be used. Submitting a confidential well report, partner data or proprietary interpretation to an external service may breach licence terms, partner agreements or regulatory restrictions, and the question of whether submitted content is retained or used for training depends on the specific service and contract. Organisations need a clear position on what may be submitted where, and technical staff need to know it, because the convenience of these tools makes casual breaches easy.
By the end of this training, participants will be able to:
- Explain how generative models operate and why they fail as they do
- Frame tasks and construct prompts that produce useful output
- Apply these models to document and report analysis
- Extract structured data from unstructured sources
- Generate and verify code for technical tasks
- Configure retrieval based approaches grounded in internal documents
- Define verification requirements proportionate to consequence
- Apply confidentiality and data handling rules
- Recognise the limitations and characteristic failure modes
- Establish governance for use in technical work
The training is practical, working on subsurface documents and data. Participants apply the tools to summarisation, extraction and code tasks and verify every output against the source, which demonstrates both the usefulness and the failure rates directly. Retrieval configurations are built against supplied document sets. Governance positions are drafted and critiqued.
Organisations sending participants to this training will:
- Apply these tools to the tasks they suit
- Verify output proportionate to its consequence
- Ground answers in internal documents rather than model training
- Avoid confidentiality breaches
- Recognise and reject invented technical content
- Establish governance for technical use
Participants will:
- Use these tools effectively for suitable tasks
- Frame requests that produce useful output
- Verify output reliably
- Build retrieval based configurations
- Recognise failure modes
- Work within confidentiality constraints
- Geoscientists and engineers in subsurface teams
- Data managers handling document holdings
- Technical staff with document heavy workloads
- Digital and innovation staff
- Subsurface team leaders setting practice
- Technical staff writing code and scripts
- Managers establishing governance
Module 1 - What These Models Are
- Language models and their operation
- Training on text and its implications
- Next token prediction
- Why output is fluent regardless of accuracy
- Invention and its mechanism
- Knowledge cut-off and currency
- Context windows and their limits
- Reasoning capability and its limits
- Numerical and quantitative weakness
- Variation between models
- Multimodal capability
- What they are not
Module 2 - Prompting and Task Framing
- Task specification clarity
- Providing context and source material
- Specifying output format
- Examples in prompts
- Step by step instruction
- Asking for sources and reasoning
- Asking for uncertainty acknowledgement
- Iterating on output
- Decomposing complex tasks
- Tasks suited to these models
- Tasks that are not
- Prompt reuse and templates
Module 3 - Document and Report Analysis
- Summarisation of technical reports
- Multi-document comparison
- Question answering over documents
- Identifying key findings and conclusions
- Extracting assumptions and caveats
- Literature and publication review support
- End of well report analysis
- Partner and regulatory document review
- Translation of technical material
- Verification against the source
- What summarisation loses
- Appropriate use in technical review
Module 4 - Data Extraction
- Structured extraction from unstructured text
- Table extraction from reports
- Extraction from scanned documents
- Optical character recognition quality effects
- Well header and completion data extraction
- Core analysis data extraction
- Test and production data extraction
- Specifying output schemas
- Confidence and flagging uncertain extractions
- Validation of extracted values
- Range and consistency checking
- Human review of extraction at scale
Module 5 - Code Generation
- Code generation for technical tasks
- Specifying requirements clearly
- Data processing scripts
- Plotting and visualisation code
- Format conversion utilities
- Debugging assistance
- Explaining unfamiliar code
- Refactoring and documentation
- Testing generated code
- Testing against known answers
- Subtle errors in plausible code
- Responsibility for generated code
Module 6 - Retrieval From Internal Documents
- Retrieval augmented approaches
- Document indexing and chunking
- Embedding and semantic search
- Passing retrieved passages to the model
- Citing sources in answers
- Reduction in invention
- Checking answers against citations
- Document set preparation
- Handling scanned and poor quality documents
- Access control in retrieval systems
- Evaluation of retrieval quality
- Building a retrieval system for a team
Module 7 - Verification and Confidentiality
- Verification proportionate to consequence
- Verification methods by task type
- Checking against source documents
- Testing generated code
- Independent calculation of numbers
- Sampling for high volume tasks
- Tasks where verification is impractical
- Confidentiality of submitted content
- Service terms and data retention
- Partner and licence restrictions
- Regulatory and export restrictions
- Internal and self-hosted options
Module 8 - Limitations and Governance
- Characteristic failure modes
- Invented references and figures
- Confident incorrect technical statements
- Arithmetic and unit errors
- Bias in training data
- Overreliance and skill erosion
- Governance policy content
- Approved use cases
- Prohibited use cases
- Disclosure of use in deliverables
- Training and competence requirements
- Reviewing and updating policy
Upon successful completion of this training course, delegates will be awarded an official Certificate of Completion issued by the Petroleum Engineers Association (PEA), an ISO 9001:2015 certified training organization. The certificate carries 10 Credits and formally records the total learning hours completed.
Each certificate is signed by the Course Facilitator and the CEO of the Petroleum Engineers Association, and serves as verifiable proof of professional training that delegates can present to employers and professional bodies worldwide.
Frequently Asked Questions
All course bookings made through PEA are strictly non-refundable. By registering for a course, you acknowledge and accept that all fees are payable in full and are not subject to refund under any circumstances, including changes in personal or professional commitments or partial attendance.
PEA reserves the right to make reasonable adjustments to course content, trainers, or schedules where necessary, without entitling delegates to a refund. Comprehensive details of each course — including objectives, target audience, and content — are clearly outlined before enrolment, and it is the responsibility of the delegate to ensure the course's suitability prior to booking.
For any inquiries related to cancellations or bookings, please contact our support team, who will be happy to assist you.
