Summer 2027 AI Applied Research Internship
$52,000–$52,000 year
On-siteWashington, United States
Job Summary
Formulate reinforcement learning and optimization models to sequence nuclear construction across multiple sites, allocate capital under uncertainty, and secure distributed infrastructure. Build simulation environments that faithfully represent operational processes, then rigorously evaluate and iterate models before deploying them into decision systems. Translate messy real-world challenges into tractable mathematical formulations, design reproducible experiments, and collaborate with engineering teams to ensure safe model monitoring and updates. Work alongside nuclear industry experts to deliver solutions that inform real operational decisions and create business value.
Required Qualifications
- Currently pursuing an MS or PhD in Computer Science, Machine Learning, Operations Research, Applied Math, Economics, Statistics, or a related quantitative field
- Returning to your MS or PhD program after the fellowship (expected graduation December 2027 or later)
- Production-quality Python and PyTorch, with solid machine learning fundamentals
- Hands-on experience (coursework, research, or projects) with at least one of: reinforcement learning, mathematical optimization, simulation and modeling, or time-series forecasting
- Able to translate a messy real-world process into a tractable formulation (an MDP with sensible state, action, and reward, or an optimization model) and explain the modeling choice
- Demonstrated ability to design, implement, and evaluate experiments, with reproducible research practices (version control, testing)
- U.S. Person status (U.S. citizen or lawful permanent resident)
- Willing and able to work on-site in Washington DC, five days a week, for the full 12-week program
Desired Qualifications
- Deep RL: policy gradient (PPO, SAC) or value-based (DQN, IQL) methods; offline / batch RL (CQL, IQL, TD3+BC, Decision Transformer)
- Combinatorial optimization with ML: graph neural networks for scheduling or routing, or neural combinatorial optimization
- Multi-agent RL (MAPPO, QMIX) or stochastic / robust optimization (CVaR-constrained, chance-constrained, distributionally robust)
- Uncertainty quantification; a game-theory or behavioral-science perspective on decision-making
- MLOps for models in production: serving, monitoring, retraining, and distribution-shift detection
- Domain exposure: construction or infrastructure operations, energy or electricity markets, industrial control systems, or critical-infrastructure security
Hiring someone like this?
Get your role in front of qualified candidates on Sorce.