Every operator in the oil and gas industry knows the name WellView. As Peloton's well information management system, it is the digital backbone that centralizes data across a well's entire lifecycle—from initial planning through drilling, completions, and eventual abandonment.
It's designed to be the single, reliable source of truth for managing the immense complexity, high costs, and significant risks inherent in drilling operations.
In essence, systems like WellView are indispensable for:
- Standardizing data for more reliable decision-making.
- Preserving critical operational knowledge over decades.
- Facilitating regulatory reporting and compliance.
The Challenge: A Treasure Trove Trapped in Fragmentation
Despite its critical importance, WellView data often presents a massive challenge for the very engineers who need it most. The information is highly technical, fragmented across complex relational tables, and stored in overwhelming volumes—think drilling logs recorded every foot or every 15 minutes.
Engineers spend an inordinate amount of time acting as data detectives, manually piecing together facts from databases, reports, and unstructured field comments. This process is slow, heavily reliant on the experience of senior staff, and creates a significant bottleneck.
While existing dashboards show numbers and graphs, they often miss the critical operational insights buried in textual reports and narrative comments from the field. This fragmentation slows decision-making and limits the use of valuable, hard-won operational knowledge.
A National Operator's Problem: The Time Log Tangle
This wasn't a hypothetical problem; it was the exact challenge faced by a major National Oil Company we partnered with. With over 1,000 wells worth of data in their system, they were particularly keen to gain quick insights from the time logs.
These time logs are the narrative heart of the operation—a meticulous, step-by-step account of everything that happened during drilling, as recorded by the crew in the field. While the structured data gives an overview, it’s the time log entries that provide context and explain why certain events occurred. For our client, this amounted to approximately 750,000 separate time log instances—a mountain of unstructured text too large for any human to manually sift through.
The Solution? What if you could simply ask Wellview a question in plain English and get an immediate, consolidated answer? This is where Generative AI comes in.
The Data Strategy: Building a Foundation, Not Just a Pilot
A Generative AI search pilot is only as good as the data it's trained on. For our Wellview GenAI Search, we knew that taking shortcuts on the data modeling side would quickly limit the Agent's performance and future scalability.
Instead of a quick-and-dirty model just for the PoC, we designed a strategy to build a future-proof foundation..
Here is a breakdown of our three-step approach:
1. Starting with the Source: NEAT and the Source Data Model
Before we could map data, we first had to understand it. We used NEAT (Cognite's modeling tool) to create a Source Data Model directly from the Wellview data. This step ensures that we have a precise, well-defined representation of the raw data as it exists today, setting the stage for everything that follows.
2. The Domain Core: Architecting for the Future
The heart of our strategy is the move from the Source Model to a Domain Data Model. Specifically, we adopted the industry-standard CDM Wells model—an architecture developed by our team and grounded in Common Data Model (CDM) principles.
Why use an "overkill" model for a pilot? Because our goal extends beyond just this PoC. This CDM Wells model is designed to incorporate nearly all potential data sources for Drilling and Wells (D&W), aligning with our client's long-term vision.
- The Bulk of the Work: Anticipating this mapping and population step to be the most time-consuming part of the data modeling phase, we focused on meticulous, scalable execution to ensure data fidelity.
3. The Solution Layer: Optimizing for the GenAI Agent
Once the Domain Data Model was robustly populated, we created the Solution Model(s) necessary for the GenAI Agent. This final step is crucial for performance and user experience:
- Readability for the Agent: We carefully renamed and described all properties to be clear, unambiguous, and easily understandable by the language model (Agent).
- Targeted Efficiency: While the Domain Model is comprehensive, the Solution Model is lightweight and optimized, containing only the data necessary for the Agent to accurately handle the user's questions.
By following this three-layer approach, we're not just creating a Gen AI pilot; we're establishing the strategic data architecture required to scale Wellview insights across the entire enterprise.
Building the Intelligence: Atlas AI and the Query Knowledge Graph
With the data successfully mapped and harmonized into the Wells Core Model—and its GenAI-friendly Solution Layer—the next challenge was connecting the power of a Large Language Model (LLM) to this complex industrial structure. We needed a translator that could take a plain English question and transform it into a highly specific data query.
This is where Cognite Atlas AI™ and its Query Knowledge Graph functionality became central to the pilot.
Atlas AI: The Agent Workbench
We used the Atlas AI industrial agent workbench to build, manage, and orchestrate the Gen AI agent. Atlas AI extends the capabilities of Cognite Data Fusion (CDF), giving the LLM secure, real-time access to the data structure we meticulously built.
The Query Knowledge Graph (QKG): The Crucial Translator
The Query Knowledge Graph is the critical bridge. It operates by understanding the semantic structure of our data model: the Wells Core Model and its simplified Solution Model.
The process works as follows:
- User Input: An engineer asks a natural language question (e.g., "What was the average ROP for Well X, and why was there a delay on June 15th?").
- Semantic Mapping: The Atlas AI Agent, using QKG, analyzes the question against the renamed and described properties in our Solution Model.
- Query Generation: QKG automatically translates the natural language request into a precise, efficient graph query that targets the specific data instances within the Wells Core Model (e.g., retrieving the depth time series data and the specific timelog entry for the delay date).
- Data Retrieval & Grounding: The query executes, retrieving only the necessary data and the unstructured time log text.
- LLM Synthesis: This grounded data (numbers, metrics, and narrative context from the 750,000 time log entries) is passed to the LLM for summarization.
Accuracy and Context
By using QKG, we achieve two critical outcomes:
- Precision: The LLM does not hallucinate because it doesn't try to guess the answer. Instead, it relies entirely on the precise data retrieved by the graph query—a form of Retrieval-Augmented Generation (RAG) native to the industrial knowledge graph.
- Depth: The agent can synthesize the highly structured metrics (like average depth) with the unstructured, contextual details (like the reason for the delay found in the time log entry text), delivering a comprehensive and actionable answer that traditional dashboards could never provide.
This approach allowed us to move from complex, fragmented data to a sophisticated, context-aware search agent quickly and securely.
Results: Unlocking Actionable Insights on Demand
The successful deployment of the Gen AI Agent—powered by the foundational Wells Core Model and the Query Knowledge Graph—immediately transformed how engineers interact with Wellview data. Our pilot demonstrated a shift from time-consuming, manual data detective work to instant, precise, and actionable insights.
The true measure of success lies in the types of questions our engineers can now answer, questions that were previously impossible to answer in a single, consolidated search.
Beyond Dashboards: Deep, Contextual Querying
The GenAI Agent excels at bridging structured numerical data with the critical, unstructured text buried in the 750,000 time log entries. This capability provides a level of context essential for true operational decision-making.
| Insight Category | Example Queries Enabled by GenAI Search | Key Benefit |
|---|---|---|
| Operational & Safety Review | "Identify the 10 most common type of safety observations recorded in the past three months on rigs 1, 2, 3,4 & 5." | Proactive Safety: Quickly identifying trends across assets to prevent future incidents. |
| Historical Performance | "Show me the last five wells where a stuck pipe incident was recorded." or "Generate a summary of all the BHA failures over last 5 years with depth of failure, component that failed, and NPT incurred." | Failure Analysis: Instant access to aggregated historical component performance for better planning. |
| Drilling Diagnostics | "Find all wells where key seating was observed. Summarize the indications related to key seating." | Troubleshooting: Rapidly correlating common symptoms and events across the entire well fleet. |
| Contextual Reporting | "Give a summary of the 9-5/8" cement jobs done in B platform wells. Include volumes and pressures as recorded in the time log for each step of the operations." | Actionable Detail: Fusing structured (volumes, pressures) and unstructured (time log step descriptions) data for high-fidelity reporting. |
| Event Root Cause | "Identify wells where a well control event occurred. Capture the operation at the time and how was the event resolved (i.e., drillers methods, bull heading)?" | Learning from Incidents: Extracting the exact operational context and resolution strategy from narrative text. |
| Financial Analysis | "Provide a breakdown of well costs on X field, comparing estimated vs. actual expenditures." | Financial Clarity: Consolidating operational and financial data for cost control. |
| Parameter Comparison | "How does the final drilling parameters compare between Well X and Well Y at their 8.5" TDs?" | Benchmark Analysis: Comparing specific engineering outcomes across different wells quickly and reliably. |
The Value Unlocked
The GenAI Agent effectively acts as an always-on expert analyst, capable of processing vast quantities of data in seconds. It converts ambiguity (a natural language question about an operation) into certainty (a precise, fact-grounded answer), dramatically reducing the time engineers spend searching and maximizing the time they spend acting on critical insights.
This capability, especially in leveraging the approximately 750,000 timelog instances, moves valuable operational knowledge from a dormant archive to an active, strategic asset.
From Pilot to Production: The Time-to-Value Advantage
The real strategic takeaway from this GenAI Search Pilot is not just the successful answering of complex queries, but the speed at which we can now deploy it at scale.
By investing the time upfront to establish the robust, forward-looking CDM Wells Core Model and integrating it with the Cognite Atlas AI™ Query Knowledge Graph, we have effectively solved the hardest part of the problem. That effort now translates into a dramatic acceleration for future deployments:
Instant Value, Accelerated Delivery
Once the foundational model is in place—a model we now possess—the process of extending this GenAI Search capability to new assets or even new clients becomes remarkably fast.
Instead of months spent on bespoke data modeling and integration, we can now achieve full deployment in days to weeks, depending only on the volume of well data to be ingested and mapped. This rapid time-to-value means your drilling engineers can realize instant benefits:
- Time Saved: Engineers are immediately freed from laborious, manual data aggregation and report-pulling tasks, allowing them to focus on high-value engineering analysis.
- Hidden Insights Unlocked: Critical operational knowledge, previously locked in hundreds of thousands of unstructured time log entries, is made instantly searchable. Insights that could have been completely missed through manual processes—such as subtle correlations between specific equipment failures and operational narratives—are now brought directly to the engineer’s screen.
- Knowledge Democratization: The solution effectively democratizes decades of operational expertise. Every engineer, regardless of experience level, can access consolidated, trustworthy knowledge by simply asking a question.
By transforming your complex Wellview data into a conversational knowledge source, the Cognite GenAI Search Agent is no longer a pilot—it is an essential tool for operational efficiency, risk mitigation, and continuous improvement in your drilling organization.