Applied AI Scientist - Hybrid

XPO•Boston, MA

1d•$100,000 - $120,000•Hybrid

About The Position

The Applied AI Scientist role at XPO focuses on designing and building agentic experimentation layers over optimization models, developing evaluation harnesses for benchmarking, and implementing operational safeguards for autonomous systems. This position involves integrating modern LLM-based and foundation model architectures for forecasting tasks, collaborating with optimization/OR scientists, and communicating technical concepts to diverse audiences. The role also requires staying current with advancements in agentic systems, time-series foundation models, and applied Generative AI.

Requirements

Bachelor's degree or equivalent related work or military experience
1 year of experience designing evaluation harnesses or benchmarks to rigorously assess model or agent performance against existing baselines
Hands-on experience building applied AI systems, including one or more of: agent-based/agentic systems, experimentation frameworks, or applying LLM-based/foundation model architectures to time-series forecasting problems
Proficiency in Python and modern ML/AI frameworks and platforms (e.g. PyTorch, HuggingFace)
Strong communication skills, with the ability to explain AI system behavior and tradeoffs to technical and business stakeholders, and to collaborate closely with optimization/OR scientists on what constitutes a meaningful model improvement

Nice To Haves

Bachelor's degree in Computer Science, AI, Data Science, Engineering, or related field, or equivalent related work or military experience
Master's degree or PhD in Computer Science, AI, Machine Learning, Statistics, or related field
2+ years of experience building agentic systems for production use cases and/or R&D applications
Experience designing operational safeguards (e.g., automated checks against regressions, runaway compute, or unvalidated models reaching production) for agent-based systems
Practical experience applying time-series or tabular foundation models (e.g., Chronos or similar) to forecasting problems such as ETA prediction or demand forecasting
Practical experience with foundation model fine-tuning or post-training techniques
Practical experience applying reinforcement learning (e.g., RLHF, or RL for agent behavior and decision-making)
Experience building retrieval-augmented generation (RAG) systems is a plus
Experience applying agentic or applied AI techniques to logistics, transportation, or operations research domains

Responsibilities

Design and build an agentic experimentation layer over optimization models developed by the team's OR/data scientists, including proposing variants, running evaluations, and surfacing promising results.
Build evaluation harnesses that rigorously and automatically benchmark model and agent performance against existing baselines before promotion to production.
Implement operational safeguards for autonomous experimentation systems, such as automated regression checks, compute/cost limits, and human-in-the-loop gates before production promotion.
Evaluate and integrate modern LLM-based and foundation model architectures (e.g., Chronos-style time-series models) for ETA prediction and demand forecasting for pickup prediction.
Partner closely with the team's optimization/OR scientists to understand model internals, solver behavior, and what constitutes a meaningful improvement for P&D use cases.
Partner with machine learning engineers on the underlying infrastructure needed to run automated experimentation and evaluation at scale.
Communicate technical approaches and tradeoffs to both technical and business audiences.
Stay current on advances in agentic systems, time-series foundation models, and applied GenAI to guide adoption at XPO.