AI Research Scientist (United States, Remote)

AI Research Jobs in the United States (Remote, Full-Time)

You will run applied AI research projects for US-based customers via Rex.zone, translating open-ended research questions into measurable experiments across LLM evaluation, RLHF data design, prompt evaluation, and model performance improvement.

What You Will Do

* Own end-to-end applied research cycles: problem framing, baselines, ablations, and reporting

* Build and evaluate LLM systems using offline metrics and human-in-the-loop evaluation

* Design RLHF workflows: preference data specs, rater instructions, prompt sets, and rubric-based grading

* Create evaluation datasets and test suites: prompt evaluation, red-teaming prompts, and content safety labeling protocols

* Collaborate with data labeling teams on taxonomy, edge-case coverage, and training data quality

* Perform error analysis and model debugging to improve robustness, safety, and helpfulness

* Document methodology and results for reproducibility and auditability

Required Qualifications

* Mid-Senior experience delivering applied ML research or productionized ML evaluation

* Strong Python skills; experience with PyTorch (or similar)

* Hands-on LLM evaluation, prompt evaluation, or RLHF experience

* Experiment design, metrics selection, and statistically sound interpretation

* Familiarity with dataset development: data labeling, QA evaluation, and guideline compliance checks

* Strong written communication for research artifacts and cross-functional alignment

Preferred Qualifications

* RAG/NER/structured output evaluation experience

* Exposure to computer vision or multimodal evaluation

* Content safety labeling taxonomies and policy-aligned rubrics

* MLOps for evaluation pipelines, dataset versioning, and reproducible runs

Remote Work and Collaboration

Remote, FULL_TIME role supporting United States-based projects with distributed teams across research, engineering, and data operations.

Compensation

Hourly base pay range: $30–$50/hr.