AI Researcher

Varick Agents•San Francisco, CA
•$200,000 - $250,000•Onsite

About The Position

Varick is seeking an AI Researcher to improve AI agents’ performance on real enterprise workflows. This is an applied, production-facing research role focused on owning the post-training loop: converting data from deployments into models and evaluations that improve accuracy in repeated real-world use. The research results are deployed to agents operating inside large enterprises. The role focuses on models rather than being embedded at client sites.

Requirements

  • Strong applied machine-learning background with hands-on LLM post-training experience, such as SFT, RLHF, DPO, or similar.
  • At least one model or system taken to real users, with a clear account of its production behavior.
  • Fluency with the modern training stack and a working grasp of inference economics.
  • Disciplined evaluation practices, relying on measured results.
  • Based in San Francisco or willing to relocate.
  • Work onsite at the Financial District office five days a week.
  • Weekend work is expected.
  • Demonstrate ownership, urgency, low ego, agency, curiosity, high standards, adaptability to a startup environment, receptiveness to feedback, and a competitive mindset.

Nice To Haves

  • Published research or open-source work in LLMs and agents.
  • Experience with agentic evaluations, tool-use training, or long-context work.
  • Experience with reinforcement-learning environments.
  • Experience taking research into production.

Responsibilities

  • Own fine-tuning, preference optimization, and post-training pipelines to improve agent performance on enterprise workflows.
  • Build evaluation suites and golden sets to measure whether changes improve performance before deployment to clients.
  • Turn production traces and failure cases into training data and targeted improvements.
  • Work with the platform team on the model layer, including routing, distillation, and cost and latency budgets.
  • Assess when fine-tuning is more effective than prompting, retrieval, or scaffolding.

Benefits

  • Meaningful equity at a stage where it still compounds.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service