KEY ACCOUNTABILITIES

● Build ML solutions for decision-making problems: planning, sequencing, routing, 
allocation, and resource utilization. 
● Prototype fast using agentic coding tools (e.g., Claude Code-style workflows): 
generate scaffolds, refactor, write tests, iterate on experiments—while maintaining 
strong engineering discipline. 
● Develop and evaluate models in areas like: 
    ○ Optimization & solvers: MILP/CP-SAT, heuristics/metaheuristics, constraint 
programming, search methods 
    ○ Deep RL / Decision Intelligence: RL baselines, offline RL, bandits, 
MCTS-style planning, policy/value learning 
    ○ Predictive ML: forecasting and estimation models that feed decision systems 
● Design robust evaluation harnesses: offline simulation, counterfactual testing, 
ablations, and scenario analysis; define KPIs and acceptance thresholds. 
● Collaborate with ML engineers to support productionization: latency/throughput 
constraints, monitoring, reproducibility, model versioning, and safe rollout. 
● Write clear technical documentation and communicate findings to both technical and 
non-technical stakeholders. 

What We’re Looking For (Required) 
● 0–5 years experience in applied ML / data science / applied research (internships, 
thesis work, and strong project portfolios count). 
● Demonstrated experience using agentic coding assistants in real development 
(e.g., Claude Code, similar agentic coding environments) to accelerate 
iteration—without sacrificing code quality. 
● Strong Python skills and comfort with ML tooling (PyTorch preferred; TensorFlow ok). 
● Solid foundations in algorithms, probability/statistics, and experimental design. 
● Ability to translate messy real-world problems into clear formulations and measurable 
success metrics. 

Strong Plus / Preferred 
● Prior work in Deep RL (a strong differentiator), such as: 
○ PPO/SAC/DQN style methods, offline RL, imitation learning, MCTS/planning 
hybrids 
○ Building environments/simulators, reward design, stability/debugging, 
evaluation 
● Experience with simulation-based evaluation or digital twins (even lightweight 
simulators). 
● Familiarity with MLOps basics: MLflow, Docker, CI/CD, model monitoring. 
● Domain exposure to logistics/supply chain/industrial operations (nice-to-have, not 
required). 

Tools & Tech (Indicative) 
Python, PyTorch, OR-Tools / solver stacks, RL libraries (Ray RLlib / Stable Baselines), SQL, 
Docker, Git, MLflow; cloud platforms a plus. 

#LI-MP1