This webinar was about utilizing generative AI in analyzing subsurface geology. It involved presenters from the subsurface analytics software company Spotfire. It is most relevant for subsurface geologists and engineers with access to large datasets and IT budgets.
They utilize vibe coding of
energy data, which refers to coding with AI assistance. Their focus is on “the
last mile,” which refers to the final steps before the AI-enhanced energy data
can be used.
It was noted that AI needs to
know how to ask the right questions. The answers are in the data, and the right
questions can bring them out.
It was noted that the
‘Insight vs. Decision Problem’ is an industry reality. The bottleneck is moving
from access to interpretation and actionability.
They say there is an AI
implementation gap. Limited trust and fragmentation are barriers.
Domain-aware analytics are
needed, which speak the language of geologists and engineers. Thus, the
Spotfire approach involves industry-native AI, GenAI for energy, but there are
challenges. Wrong answers and hallucinations are unacceptable. AI should be
designed to be a complement.
Spotfire is a subsurface
analytics platform that integrates the subsurface with architecture. It is
designed to be a force multiplier, not a force replacer. LLM can search
databases, suggest visuals, and recommend charts. It can work through
chatbot-based features as well. Prompting is required to manage AI. The expert
needs to stay in the loop.
Technical Deep Dive
Spotfire is the management
platform for subsurface data. LLMs are provided by other platforms such as
OpenAI, Azure, Claude, etc.
One can do inquiries within
Spotfire, rather than in the main AI platforms. Data can be written for
functions such as log analysis. Input/output parameters must be set up, which
is what Spotfire does. There is no need to learn Python. Spotfire Copilot 2.3
is the latest release. It has a strong agentic component. Agents can look at
daily drilling reports, log analysis, production decline, and analysis, etc.
Demo Act 1: Daily Drilling Report Agent – agents analyze
data in the reports, make it searchable, and can answer questions. They can ask
and answer questions for multiple wells.
Demo Act 2: Well Recompletions Advisor – all data sources
can be used to pick wells to recomplete. Well log analysis is included. Wells
can be scanned and ranked as recompletion candidates.
Demo Act 3: Observability Hub – observability is enhanced.
Key Takeaways
1) Industry native
2) Architecture is the product
3) Experts build, not just use.
Q&A – the hard part now
is turning data into decisions faster. Q: Quality assurance? – A:
Within Spotfire, you can see how the LLMs are working with the data. Citations
are generated. Hallucinations can be discovered/reduced. One can monitor and
audit agent responses within Spotfire. Geology and petrophysical workflows can
be generated in Spotfire.
AAPG’s Webinar Summary
“This webinar, hosted by Susan Nash AAPG, featured a
technical deep dive into generative AI implementation for subsurface analytics
with presenters Alessandro Kimera, Drew Scherer, and Athir Alatar from
Spotfire. The session focused on architecture and frameworks needed to move AI
from isolated pilots to scalable enterprise-ready solutions, introducing the
concept of "vibe coding" which allows domain experts to build and
customize data applications through AI conversations. The presenters demonstrated
two key AI agents: a Daily Drilling Reports (DDR) agent that processes
unstructured drilling data to create searchable knowledge databases, and a Well
Recompletions Advisor that scores well candidates for re-entry decisions by
analyzing multiple data sources including completion data, geological
information, and production history. They emphasized that successful AI
implementation requires industry-native, expert-supervised systems with strong
governance and observability, as generic tools fail to understand domain-specific
data like well logs and decline curves. The webinar also highlighted customer
success stories including Liberty's analysis of 50 billion operational data
points and S&P Global's 80% time reduction for well analysis, concluding
with a preview of upcoming Webinar 3 which will focus on Agentic AI for energy
workflows.”
No comments:
Post a Comment