• Traditional age estimation of rocks by biostratigraphers is time-consuming and subjective
• Client wanted to automate the process using advanced analytics
• Needed a scalable solution to improve consistency and speed in age prediction

• Expanded fossil species lookup from 100s to 60,000 to train the AI model
• Built a digital twin of the biostratigraphy process using historical well data (depth + fossil IDs)
• Trained AI engine to learn from historical patterns and predict age across depth continuously

• 10x faster processing speed; 90% of samples no longer need manual review
• Age estimated continuously across depth levels
• Outputs categorized by confidence level, helping teams prioritize manual verification only where needed

• Predict daily price changes of monthly futures up to 100 days before delivery
• Needed to understand complex price drivers in European gas markets: demand, supply, weather, geopolitical events, and financial indicators
• Goal: build a forecasting model as a decision support tool for the trading team

• Analyzed historical gas prices, weather, macro indicators, Bloomberg news, forward curves, and more
• Combined structured and unstructured data to train a model predicting the expected percentage change in gas futures contracts
• Forecasted 100 days ahead, supporting traders with high-confidence insights

• Achieved 24% higher accuracy than benchmark models
• Enabled better-informed buy/sell decisions in futures markets
• Supported traders with early warning signals and optimized risk-reward positions

• Estimate hydrocarbon saturation levels across large sedimentary basins with hundreds to thousands of wellbores
• Traditional methods required manual expert analysis, which was time-consuming
• Client wanted to save time and man hours by automating the saturation estimation process

• Analyzed hundreds of wireline logs (resistivity, porosity, gamma ray) and supporting datasets (casing, formation group, geolocation)
• Used explainable and predictive variables to estimate saturation
• Built a model that imputes missing values in wireline logs, reducing need for expert review
• In 12 weeks, addressed data gaps and predicted missing values for key logs

• 2 FTE years saved through automation of the saturation estimation process
• Enabled client to effectively scan entire basin for hydrocarbon signals and reprioritize Geo-Scientist time to focus on top-ranked prospects
• Achieved 20% higher accuracy compared to traditional expert-based methods

• Client aimed to increase oil rate by improving forecasts of:
• Water Cut (percentage of water in extracted fluid)
• Gas/Oil Ratio (GOR) to understand volume of oil vs. gas‍
• Allocated Rates to determine oil per sensor per branch

Used historical well test data, sensor readings, and allocated rates. Built a model through:

• Data preprocessing (removing outliers, time alignment, lookups)
• Feature engineering and iterative learning
• Trained prediction model using SparkBeyond platform
• Analyzed 3.9 billion explainable features

• 28% higher accuracy compared to baseline
• RMSE reduced from 9.2 to 6.6
• More precise oil production estimation

• Predicting gas demand within-day, 1 and 2 days ahead
• Consumption by industrial clients often fluctuates unpredictably, causing supply imbalances
• Needed to build a robust forecasting model to support operations and trading with better accuracy

• Used historical offtake, metering data, weather data and client metadata
• Developed model to predict actual gas consumption more accurately than clients’ nominated submissions
• Model tailored to two major European markets

• 6.5% higher accuracy compared to benchmark models
• Forecasts provided more reliable input for trading decisions
• Enabled proactive planning and reduced last-minute balancing costs