# Lead/Senior Research Scientist (Agentic RL) at Jobgether

> Remote · United States · Staff · ML / AI · open. Source: https://alderhq.app/jobs/lever%3Ajobgether%3A5b30bb2a-1c98-44a4-b68a-8f0a0eb5cf0a (read 2026-11-21).

## Facts (observed)

- Company: [Jobgether](https://alderhq.app/companies/jobgether)
- Location text: US
- Where it can be done from (inferred): Remote · United States (evidence: "Location: US")
- Posted: 2026-11-19 (2 days ago)
- First seen by alder: 2026-11-19 · last verified: 2026-11-20
- Reposted: 0 times · edited: 0 times
- Apply on the employer's site: https://jobs.lever.co/jobgether/5b30bb2a-1c98-44a4-b68a-8f0a0eb5cf0a
- Skills listed (inferred): Python, PyTorch, LLMs

## Pay and sponsorship

- Posted pay: not posted
- Market band for similar posted roles (United States, n 389): p25 $220k, median $255k, p75 $281k

## Description (the employer's text)

### Accountabilities

Design and develop benchmarks, training datasets, reward models, and training signals that improve model performance across complex professional workflows.

Develop methods for tasks where ground truth is limited, delayed, subjective, or difficult to verify.

Conduct rigorous research experiments and diagnostics, including ablation studies, to distinguish algorithmic improvements from data, reward, and system-related effects.

Translate experimental findings into measurable improvements in production models and model development processes.

Build and expand model evaluation suites to identify strengths, limitations, and performance gaps across professional use cases.

Use evaluation results to inform training objectives, research priorities, and technical trade-offs.

Research and develop agentic reinforcement learning methods that improve planning, long-context reasoning, tool calling, multi-turn interactions, and long-horizon task execution within real agent environments.

Contribute to post-training approaches, including policy optimization, reward modeling, and the design of reliable training signals for agentic behavior.

Contribute to continued pre-training and post-training of open-weight models, including Direct Preference Optimization (DPO), reinforcement learning, training data design, and strategies for mitigating forgetting through replay.

Develop evaluation methods for complex legal, tax, and other professional workflows, including expert-annotated benchmarks, LLM-as-a-judge systems, evaluation rubrics, and human preference studies.

Work closely with Research Engineers to develop and improve large-scale model training and evaluation infrastructure.

Collaborate with legal, tax, and other professional subject-matter experts to ensure model training and evaluation reflect real-world requirements and domain-specific expectations.

Partner with product teams and other model users to understand adoption needs and support the integration of research outcomes into practical applications.

Publish research regularly at leading machine learning and natural language processing conferences and contribute to the broader research community.

At Lead level, define and own a research agenda, translate emerging research methods into a strategic roadmap, and mentor scientists and engineers.

Depending on specialization, contribute primarily to one or two of the following areas: agentic reinforcement learning, model training, or model evaluation.

### Requirements

Education: PhD in Machine Learning, Computer Science, or a related discipline, or equivalent practical research experience.

Research and development experience: Hands-on experience in LLM post-training and/or model evaluation, preferably at the scale of frontier model development.

Technical skills: Strong Python programming skills and practical experience with PyTorch.

Research impact: Demonstrated research contributions through publications at leading venues such as NeurIPS, ICML, ICLR, ACL, EMNLP, or COLM, and/or direct contributions that have influenced an LLM development cycle.

Cross-functional collaboration: Experience working with researchers, engineers, and subject-matter experts to turn research ideas into practical training datasets, evaluation frameworks, or production capabilities.

Communication: Excellent written and verbal communication skills, with the ability to explain complex research findings and collaborate effectively across disciplines.

Adaptability: Comfort working in a fast-paced, evolving research environment where problems may be ambiguous and priorities can change.

Lead-level requirements: For Lead positions, at least three years of research experience after completing a PhD, a track record of leading research agendas through to published or deployed outcomes, and experience mentoring or supervising researchers.

Preferred reinforcement learning frameworks: Experience with RL post-training frameworks such as NeMo-RL, verl, OpenRLHF, or TRL.

Preferred evaluation frameworks: Familiarity with tools such as Inspect or lm-evaluation-harness.

### How Jobgether works

We use an AI-powered matching process to ensure your application is reviewed quickly, objectively, and fairly against the role's core requirements. Our system identifies the top-fitting candidates, and this shortlist is then shared directly with the hiring company. The final decision and next steps (interviews, assessments) are managed by their internal team.

We appreciate your interest and wish you the best!

Why Apply Through Jobgether?

Data Privacy Notice: By submitting your application, you acknowledge that Jobgether will process your personal data to evaluate your candidacy and share relevant information with the hiring employer. This processing is based on legitimate interest and pre-contractual measures under applicable data protection laws (including GDPR). You may exercise your rights (access, rectification, erasure, objection) at any time.

#LI-CL1

---
Data: alder (https://alderhq.app). Cite https://alderhq.app/jobs/lever%3Ajobgether%3A5b30bb2a-1c98-44a4-b68a-8f0a0eb5cf0a and the date you read it. Observed facts come from the employer's own listing; location reach, level, family and skills are inferences.
