Job Description
Research Engineer – Post Training Reinforcement Learning
Location: San Francisco (Hybrid)
About TensorStax
TensorStax is building fully autonomous AI systems to manage and maintain mission-critical data infrastructure and pipelines. We leverage reinforcement learning to enhance language models' ability to reason over large-scale data lakes and warehouses, detect pipeline failures, construct new pipelines with high precision, and enable agentic behavior—allowing systems to proactively identify and resolve issues autonomously.
What You’ll Do
As a Research Engineer specializing in Reinforcement Learning, you will:
- Develop and refine reward functions to optimize agent behavior for complex data engineering tasks.
- Create RL gym environments for language model agents.
- Fine-tune language models using reinforcement learning techniques such as PPO, DPO, and KTO.
- Stay at the forefront of research on RL for language models, incorporating advancements like GRPO, SWE-Gym, and SWE-RL into practical applications.
- Curate and build high-quality datasets for supervised fine-tuning (SFT) and RLHF.
- Design experiments to evaluate and improve the agentic capabilities of language models in data environments.
What We’re Looking For
- Deep understanding of reinforcement learning, reward shaping, and optimization strategies.
- Strong familiarity with LLM fine-tuning techniques (PPO, DPO, KTO) and their applications in reinforcement learning.
- Knowledge of recent advancements in RL for language models (GRPO, SWE-Gym, SWE-RL).
- Experience curating and constructing high-quality datasets for fine-tuning.
- Strong problem-solving skills and a history of working on complex ML projects.
- High agency—ability to work independently, experiment proactively, and drive research initiatives forward.
Bonus Points
- Experience with distributed training in PyTorch (DDP, FSDP).
- Hands-on experience designing RL environments for traditional RL problems.
- Contributions to open-source projects in RL, LLMs, or ML infrastructure.
- Familiarity with data lakes and warehouses (Snowflake, BigQuery, Redshift).
Benefits
- 100% employer-covered health, dental, and vision insurance.
- 401(k) with company match.
- Access to Bay Club or Equinox in San Francisco.