Zanskar1mo ago

Senior Machine Learning Scientist

Salt Lake CityFull Timesenior

Data ScienceMachine Learning ScientistData & AI

0 views0 saves0 applied

Apply Now

Quick Summary

Key Responsibilities

POMDP-style planning and belief-space decision makin

Technical Tools

Data ScienceMachine Learning ScientistData & AI

Role Overview

Title: Senior Machine Learning Scientist (Surrogate modeling & decision science in the earth sciences)

Hours: Full-Time, Salaried

Location: Salt Lake City, UT, Hybrid (3 days in office, 2 days can be remote)

Benefits Eligible: Yes

Manager: Head of Reservoir R&D

Why we exist

Geothermal energy is the most abundant renewable energy source in the world. There is 2,300 times more energy in geothermal heat in the ground than in oil, gas, coal, and methane combined. However, historically it’s been hard to find and expensive to develop. At Zanskar, we’re building technology to find and develop new geothermal resources in order to make geothermal a cheap and vital contributor to a carbon-free electrical grid.

To do that, we combine deep subsurface expertise with advanced AI technologies—including modern machine learning, scalable scientific computing, and uncertainty-aware modeling—to dramatically improve geothermal discovery and development outcomes. We build systems that can learn from sparse and noisy data, emulate expensive physics simulations, and help teams make faster, higher-confidence decisions about where to drill and how to develop fields.

Who you are

You will help build the modeling and decision-making core of Zanskar’s geothermal exploration software. This role blends scientific machine learning (surrogate modeling) with sequential decision-making under uncertainty. A successful candidate will:

Explore: you’re open-minded about methods and will prototype, benchmark, and iterate across approaches.

Reproduce & adapt: you can implement ideas from papers and new frameworks quickly, then harden the best ones into reliable workflows.

Decision-minded: you care about end-to-end outcomes (value, risk, time-to-decision), not just model accuracy.

Uncertainty-first: you build models that are accurate, well-calibrated, and dependable under distribution shift and sparse data regimes.

Collaborative: you work well with domain experts and can translate between geology/engineering intuition and ML systems.

What you’ll do

Build fast, reliable models that emulate or augment computationally expensive physics-based simulations (e.g., reservoir, wellbore, and coupled multi-physics workflows).

Evaluate and compare multiple modeling approaches (physics-informed, operator learning, transformers, diffusion models, etc.), establishing strong baselines and selecting methods based on evidence.

Build multi-step decision systems for exploration and appraisal: POMDP-style planning and belief-space decision making to recommend exploration steps.

Translate scientific and engineering questions into well-defined learning and decision problems: inputs/outputs, constraints, boundary/initial conditions, reward/cost structure, and success metrics (e.g., expected NPV, probability of success, downside risk).

Prototype, benchmark, and iterate across approaches (POMDP solvers, RL methods, VOI-style baselines, MPC-style replanning), then harden the best ones into reliable workflows and APIs.

Collaborate deeply with geoscientists, reservoir engineers, and software engineers to integrate these models and policies into production software.

What we’re looking for

3+ years of applied ML experience, ideally in scientific ML, decision-making under uncertainty, surrogate modeling, robotics/control, or related engineering/science domains.

Expertise in python and modern ML tooling (PyTorch preferred).

Track record of taking models from prototype → rigorous evaluation → adoption by technical stakeholders.

Strong fundamentals in probability/statistics and comfort with messy, real-world scientific datasets.

Experience building or using surrogate models for expensive simulators (PDE-driven systems, multi-physics, or similar).

Relevant technical strengths

Surrogate modeling.

Sequential decision-making under uncertainty and reinforcement learning.

Software engineering: Git, code review, reproducibility, CI basics, Docker/container workflows.

Experience with diffusion models.

Exposure to subsurface modeling domains: geothermal, oil & gas, CCS, hydrogeology, geoscience, or related.

Familiarity with cloud infrastructure and data systems (SQL, object storage, orchestration).

Location and Benefits