FUJITSU RESEARCH OF AMERICA, INC. (FRA) was established in 1993 as a wholly owned subsidiary of Fujitsu Laboratories Ltd., which is a wholly owned subsidiary of Fujitsu Ltd., focusing on contributing to Fujitsu's technology value chain. FRA conducts advanced research in Artificial Intelligence, Data Science, System Components and Platform Innovations, Network and Trusted Systems Innovations, and Internet Service Innovations.
We are seeking a highly motivated PhD intern to work on LLM-based agents for end-to-end software and enterprise workflows. The internship will focus on building intelligent agents, generating high-quality trajectories, improving agent performance with training and test-time methods, and running reproducible benchmark evaluations.
The intern will collaborate with AI researchers and engineers to develop agent systems that can plan, use tools, inspect state, recover from failures, and verify their own work. The project uses Code Enterprise World Model (CEWM) style trajectories to record observation-action-next_observation transitions for SFT, RL, preference optimization, replay, and evaluation.
Job Description
- Build and evaluate LLM-based coding, browser-based, computer-use, and workflow agents.
- Improve agent performance using training data, SFT/RL methods, and test-time verification.
- Run reproducible experiments with baselines, ablations, clean reporting, and failure analysis.
- Work with agentic coding tools and AI-generated code workflows such as Cursor, Claude Code, Codex, or similar systems.
- Evaluate agents on benchmark tasks such as WebVoyager, SWE-bench, Terminal-Bench, CodeClash, or internal enterprise workflows.
Deliver results suitable for an internal research report or submission to a top AI conference.
-
Qualifications
- Currently pursuing a PhD in Computer Science, Machine Learning, Artificial Intelligence, Software Engineering, or a related field.
- Research experience with LLMs, coding agents, tool use, or agentic workflows.
- Strong Python skills; comfortable with git, debugging, unit tests, and Linux/CLI workflows.
- Experience running reproducible experiments, including baselines, ablations, and clear result reporting.
Familiarity with training or adapting LLMs using SFT, RL, DPO/RLHF methods, or trajectory data.
-
Preferred Qualifications
- Hands-on experience with coding-agent benchmarks such as SWE-bench, Terminal-Bench, WebVoyager, or similar task suites.
- Experience with inference-time verification using unit tests, learned verifiers, search, or reranking.
- Experience with multimodal agents, browser automation, vLLM, Docker/containers, CI-style evaluation, or scaling experiments.
Publication record in top AI/ML, software engineering, or systems venues.
-
Fujitsu salaries are aligned to the specific geographic location in which the work is primarily performed. It is not typical for an individual to be hired at or near the top of the range for their role and compensation decisions are dependent on the circumstances of each situation. The pay range for this role takes into account the wide range of factors that are considered in making compensation decisions including but not limited to: specific skills, qualifications, experience, and comparison to other employees already in this role. The pay range for this position is estimated at $40/hr. to $50/hr. USD.
