Project
LLM and Gen-AI Agent for Energy Industry (2024-2029)
Founder Institution
Alberta Innovates
For LLM API integration, the first objective is developing the core functionalities of the chat agent, including guiding basic tasks and answering queries. The methodology is (1) integrating the chat agent with a GPT or similar LLM API, leveraging its NLP capabilities for human-like interactions and (2) fine-tuning the LLM with subsurface energy geotechnics related data, ensuring accurate responses. The second objective is expanding chat agent’s interaction capabilities to include complex task guidance, scenario-based problem solving and predictive analytics. The methodology will be continuous learning from extensive interactions and testing of different subsurface energy development projects.
For implementation of AI-driven real-time decision support through the chat agent, the objective is utilizing the LLM’s capability to process complex data and generate predictive insights for subsurface energy development. The methodology is (1) ensuring access to the data from the modeling platform and allowing LLM in the chat agent to assist model setup and data interpretation and (2) incorporating predictive modeling and optimizations through machine learning methods.
The deliverable of this project is an LLM-powered chat agent with advanced human interactions and real-time decision support that will enhance the application of the modeling platform supported by NSERC Discovery Grant. The anticipated outcome for this two-year project is to (1) advance the R&D capacity in Alberta in both AI and energy sector through integration of LLM with subsurface energy development and (2) develop a chat agent that can train and assist HQP for challenges requiring skills in AI and subsurface energy geotechnics.
More Details
Progress Achieved
LLM API Integration
We have successfully connected the chat agent to a GPT-based (or similar) API, harnessing advanced Natural Language Processing (NLP) capabilities. This integration allows the agent to comprehend user queries and respond in a conversational, context-aware manner.
User Interface Demonstration (Screenshot Reference):
The shared screenshot illustrates the chat interface where the user is guided step-by-step in a workflow (e.g., Decline Curve Analysis). Key features include:
Step-by-Step Instructions: The agent outlines the process of identifying relevant columns, configuring optional inputs, and executing the analysis.
Contextual Prompts: Users can provide data inputs (such as production tables, forecasting parameters) and receive targeted advice or clarifications
Workflow Automation: The agent is already demonstrating the ability to streamline technical workflows, reducing manual effort and potential errors.
Expanded Interaction Capabilities (Ongoing)
Complex Task Guidance:
The agent’s knowledge base can be partially tailored with subsurface energy geotechnics data. This step ensures that the model offers more accurate and relevant answers compared to a general-purpose LLM.
Complex Task Guidance:
We have started implementing frameworks that allow the chat agent to walk users through multi-step engineering and geotechnical processes. The screenshot’s Decline Curve Analysis example is an early demonstration of this functionality.
Scenario-Based Problem Solving:
Through iterative testing with various subsurface energy development scenarios, more detailed instruction will be offered for chat agent to handle diverse conditions (e.g., varying reservoir properties, different completion strategies).
Predictive Analytics Integration:
Early prototypes incorporate machine learning methods that enable the agent to predict outcomes (e.g., production forecasts) based on input data. This capability will become more robust as more domain-specific data and user feedback are integrated.