Scientific Researchers (Multiple Positions)

Job Description

Scientific Researchers — Multiple Positions \n

Open Scientific Domains | AI Training, Evaluation, Reinforcement Learning and Simulation

About Axia \n

Axia is building a marketplace and research platform that connects specialized knowledge with AI model development. Our scientific research program turns domain expertise, experimental workflows and real-world data into evaluations, training datasets and environments for scientific AI agents.

\n

We are looking for multiple researchers with deep expertise across scientific and technical disciplines to help shape this program.

The opportunity \n

How do we determine whether an AI model can conduct meaningful scientific reasoning, design a useful experiment or solve a problem beyond its training examples?

\n

At Axia, you will help answer these questions within your area of expertise. You will identify valuable scientific workflows, establish what credible performance looks like, and develop the data, evaluations and environments that help models improve.

\n

We are building a team of researchers with complementary specialties. You should bring depth in one or more domains; you are not expected to be an expert across all of them. Your research focus and responsibilities will reflect your expertise and the problems we identify together.

Scientific domains we are interested in \n

We welcome researchers with in-depth knowledge in areas including:

\n

\n
• Biology and life sciences: molecular and cellular biology, genetics, genomics, proteomics, systems biology, microbiology, neuroscience and bioinformatics.
\n
• Chemistry: organic, inorganic, physical, analytical and computational chemistry; chemical synthesis, catalysis and molecular design.
\n
• Physics: theoretical, experimental and computational physics; condensed matter, optics, photonics, quantum science and plasma physics.
\n
• Materials science: materials discovery and characterization, polymers, metals, semiconductors, composites, nanomaterials and functional materials.
\n
• Mathematics and statistics: applied and pure mathematics, probability, statistical inference, optimization, numerical analysis and formal reasoning.
\n
• Engineering and robotics: mechanical, electrical, chemical, civil, aerospace and biomedical engineering; control systems, robotics and manufacturing.
\n
• Medicine and pharmaceutical sciences: biomedical research, pharmacology, drug discovery, translational science, epidemiology and computational medicine.
\n
• Climate, Earth and environmental sciences: atmospheric science, oceanography, geoscience, hydrology, ecology and environmental modeling.
\n
• Energy and electrochemistry: batteries, energy storage, renewable energy, power systems, nuclear science, fusion and electrochemical systems.
\n
• Agricultural and food sciences: plant science, crop research, soil science, agricultural systems and food science.
\n
• Computational science and scientific machine learning: numerical simulation, high-performance computing, multiphysics modeling, inverse problems and AI for scientific discovery.
\n

\n

This list is open. We encourage applicants from other research-intensive disciplines and from fields that span several of these areas.

What you will work on \n

Depending on your expertise and role, you will:

\n

\n
• Identify meaningful scientific problems. Find gaps in model capabilities and existing benchmarks. Prioritize workflows based on scientific value, practical demand, data access, verifiability and experimental cost.
\n
• Translate domain expertise into learning tasks. Capture hypotheses, protocols, observations, analysis code, instrument constraints, uncertainty and failed experiments. Develop tasks that require reasoning, tool use and iterative investigation.
\n
• Build rigorous scientific evaluations. Establish reference standards through measured results, validated calculations, formal checks or structured expert review. Account for uncertainty and multiple scientifically defensible approaches.
\n
• Develop training data and learning environments. Work with the AI research team on demonstrations, supervised fine-tuning data, reinforcement learning trajectories, reward signals and task curricula.
\n
• Create grounded synthetic data. Use real observations, scientific constraints and expert knowledge to guide simulations and generated tasks. Test consistency, measurement noise, validity ranges and distribution shift.
\n
• Evaluate transfer to real scientific workflows. Compare performance in simulations with independent real datasets or partner-run experiments. Identify shortcuts, invalid assumptions and cases where apparent improvements fail to transfer.
\n
• Investigate what improves models. Design controlled studies comparing data sources, expert feedback, task difficulty and training strategies. Distinguish meaningful capability gains from memorization or benchmark exploitation.
\n
• Build reproducible research resources. Produce versioned datasets, documented environments, validation notebooks and clear reports. Help contributors understand what makes scientific data suitable for training and evaluation.
\n
• Collaborate across disciplines. Adapt methods with other specialists while respecting differences in evidence, measurement and validation across fields.
\n

What you bring \n

\n
• Deep scientific expertise in at least one discipline, demonstrated through original research, experiments, publications, scientific software or validated applied work.
\n
• A PhD or equivalent research experience in a relevant field.
\n
• Strong experimental design and quantitative reasoning, including controls, uncertainty, reproducibility, confounding and the limitations of observational or simulated evidence.
\n
• Practical experience with AI evaluation, scientific machine learning, model training or post-training. The required mix will vary by position; some roles will emphasize reinforcement learning, reward design and environment development.
\n
• The ability to implement research using Python and appropriate scientific computing or machine learning tools.
\n
• Judgment about what constitutes valid evidence, when specialist review is needed and when a result requires real-world measurement.
\n
• Clear technical communication and the ability to collaborate with domain experts, AI researchers and engineers.
\n

\n

We welcome both scientific specialists who have applied AI to substantive research problems and AI researchers with demonstrated depth in a scientific discipline.

Especially relevant experience \n

Experience with scientific agents, computational modeling, reinforcement learning, evaluation design, active learning, Bayesian optimization, causal inference, synthetic data or experimental automation is particularly relevant.

\n

Your background may include an AI lab, scientific or biotechnology company, engineering organization, industrial R&D team, research institute, university or AI data company.

\n

Wet-lab or physical experimentation experience is valuable where relevant to the domain. It is not required for every position.

What early success looks like \n

We will tailor initial goals to your domain and role. A typical first research cycle could include:

\n

\n
• First 30 days: identify promising workflows, review existing benchmarks and establish data, collaborator and validation requirements.
\n
• By 60 days: develop an initial task specification, dataset or environment prototype, baseline evaluation and validation plan.
\n
• By 90 days: complete a pilot study comparing data or training strategies, assess performance against independent evidence and recommend the next research direction.
\n

\n

Physical experiments will depend on partner access and available resources. Where appropriate, independent real datasets can support validation, with conclusions limited to the evidence tested.

How we assess research quality \n

We value scientifically valid tasks, reproducible methods, reliable reward signals, well-characterized uncertainty and improvement on held-out scientific workflows.

\n

We assess simulation performance, independent real-data performance and physical experimental outcomes separately. Scientific feasibility and resource requirements matter alongside model accuracy.

What to share with us \n

Please share:

\n

\n
• Your primary scientific domain, specific areas of depth and any adjacent fields you have worked in.
\n
• A scientific research artifact that demonstrates your contribution, such as a paper, experiment, dataset, software project or technical report.
\n
• An example of AI evaluation, scientific ML, model training or reinforcement learning work, where applicable to your intended focus.
\n
• A scientific workflow you would turn into a learning environment: what a model should learn, what synthetic data could preserve, what it might miss and how you would test whether improvements transfer to real scientific work.
\n

\n

We are hiring across multiple research positions and welcome applicants whose expertise can help define new scientific research directions at Axia.

\n

\n