In this episode of Stay Sharp in Digital Engineering, hosts Juliann Grant and Jonathan Scott are joined by Prithvi Singh Chauhan, Machine Learning Engineer at Geminus AI. They explore the practical application of AI and machine learning to operational digital twins across oil and gas, manufacturing, and energy systems. The discussion covers why traditional physics models fall short in real-time operations and how combining physics-native engineering with fast machine learning solves speed, staleness, and domain silos.
Key Takeaways
Traditional physics-based digital twins are built for design and back-office engineering, making them accurate but often too slow, stale, and siloed for live operations.
Scientific AI embeds fundamental laws of physics directly into machine learning models, providing millisecond-level predictions while staying reliable under unseen operating conditions.
Multimodal AI enables systems to process varied data sources together, including sensor time series, lab reports, simulator outputs, and engineering drawings.
Self-calibrating digital twins reduce manual recalibration cycles from months to real time, preventing operational drift and enabling better decision-making.
The frontier of operational digital twins includes multi-system connected models, self-maintaining architectures, probabilistic risk outputs, pre-trained foundation physics models, and LLMs acting as conversational interfaces to orchestrate physics engines.
Key Terminology:
Scientific AI: Machine learning models with built-in physics equations (conservation of mass, energy, equations of state) rather than pure data inference.
Multimodal AI: Absorbing disparate formats, frequencies, and data streams across complex engineering workflows.
World Models: Pure observational learning versus physics-grounded models in live industrial environments.
The AI-Enabled Digital Twin Stack: A layered breakdown covering the physics layer, data abstraction layer, physics-native models, continuous calibration, and decision optimization.
The Future of Industrial AI: Connecting multi-unit constraints, models that self-detect drift, foundation models for physics, and using LLMs as orchestration interfaces.
Guest Info
Prithvi Singh Chauhan is a Machine Learning Engineer at Geminus AI, where he builds physics-native AI systems combining machine learning, high-fidelity simulation, and optimization for large-scale energy systems. His work spans assets across North America, Guyana, the Middle East, India, Malaysia, and Australia. He holds an M.Eng. in Petroleum Engineering from Texas A&M University and a B.Tech. from IIT (ISM) Dhanbad. Named SPE's 2026 Young Engineer of the Year and an inventor on four US patent applications, Prithvi has co-authored around 17 peer-reviewed papers and serves as Program Chair for SPE's Data Science and Engineering Analytics technical section.
If you enjoyed this episode, please subscribe, rate, and leave a review on Spotify, Apple Podcasts, or wherever you listen to podcasts. Share this episode with colleagues and friends working in digital engineering, energy, and industrial operations!
Podden och tillhörande omslagsbild på den här sidan tillhör
Razorleaf Corp.. Innehållet i podden är skapat av Razorleaf Corp. och inte av,
eller tillsammans med, Poddtoppen.
