Research Scientist, Reinforcement Learning & World Models (Founding Team)

Thermia is the first AI-native solution vertically built for the industrial thermal chain, discovering breakthrough fluids and co-optimizing associated device, from Molecules to industrial systems.

We are a team of machine learning scientists, computational chemists, and thermal engineers united by a single ambition: to create a fundamentally new way of designing the fluids and systems that move heat through our modern economy.

Today's approach to thermal fluid and system design is fragmented, slow, and structurally incompatible with the urgency of the energy transition. Chemists develop fluids without knowledge of the systems they will run in. Engineers design thermal hardware without optimizing the working fluid. The result is decades of R&D, billions in cost and solutions that are sub-optimal by construction.

Thermia replaces this fragmented process with an integrated, physics-grounded discovery pipeline: AI accelerates what human expertise makes possible, and experiments validate what AI generates. Our platform generates entirely novel molecular candidates and co-optimizes the thermal systems associated with each discovery, building a proprietary portfolio of qualified molecules, fine-tuned models, and experimental data with every engagement.

Our work is powering the thermal infrastructure the energy transition cannot wait for.

Place: Paris, France. Level: Entry level, Contract: Permanent, Mode: Hybrid

Snapshot We're looking for a research scientist to anchor foundation models in the physical world, designing RL strategies and world models that learn the laws of molecules, fluids, and systems, and AI surrogates to accelerate the discovery of novel PFAS free thermal fluids.

The Role As a Research Scientist on the R&D team, you will build the physics-informed models that ground Thermia's discovery stack in physical reality. You will design first-principles and data-driven strategies that teach AI models how molecules behave, how fluids flow, and how thermal systems respond. You will own the AI surrogates that turn slow, expensive simulations into fast, differentiable learning signals allowing AI agents to reason and act, across molecular and continuum scales. You will implement code, run experiments, evaluations and own results end-to-end.

Your work may involve:

• Designing world models that capture the physics of molecules, fluids, and thermal systems across both molecular and continuum scales

• Building AI surrogates (PINNs, neural operators such as FNO and DeepONet, differentiable simulators) that compress expensive physics-based simulations into fast, learnable signals

• Designing RL and model-based approaches that use these world models to discover new materials, fluid geometries, and thermal systems with specific constraints

• Developing reward functions, exploration strategies, and search algorithms (MCTS, model-based RL) tailored to scientific discovery

• Validating data-driven models against ground-truth physics and grounding them in real simulation and experimental data

• Connecting molecular and physical-scale world models into a unified multi-scale framework

• Designing evaluations and ablations that answer real questions about model fidelity and downstream usefulness

• Analyzing results carefully, including debugging and failure analysis

• Communicating results clearly through plots, writeups, and paper-ready figures

About You

• PhD in Machine Learning, Applied Mathematics, Physics, or a related field

• A research track record in RL, or world models, or scientific ML, including peer-reviewed publications

• Strong implementation ability and comfort in research codebases (PyTorch or JAX)

• A deep interest in anchoring AI in the physical world and comfortable moving between ML, physics, and simulation

• Strong communication and a bias toward clarity and honesty regarding results

• High agency: you push projects forward and take initiative

In addition, the following would be an advantage:

• Prior project, internship, or PhD work in materials, drug discovery, or computational physics

• PhD work on PINNs, neural operators, differentiable physics, or world models

• Experience with model-based RL, MCTS-based search, or scientific RL

• Hands-on experience with physics-based simulators (CFD, MD, DFT)