Member of Technical Staff – Post‑Training & Applied Research

About the role

At SF Tensor we are building the world’s fastest GPU compiler and a powerful Model Foundry platform. The role focuses on owning the entire post‑training pipeline—from data curation through fine‑tuning, reinforcement learning, evaluation, distillation, and deployment—so customers can turn datasets into specialist models in days.

Key responsibilities

• Own the end‑to‑end post‑training pipeline, including data curation, SFT, preference optimization, RL, evaluation, distillation, and deployment.
• Design reward functions and RL loops in collaboration with domain experts.
• Build a trustworthy evaluation harness for rapid ship/no‑ship decisions.
• Structure and generate datasets, including synthetic data pipelines.
• Distill specialist models into smaller, edge‑deployable versions.
• Identify and eliminate critical path bottlenecks to accelerate time‑to‑model.
• Embed with customers as a forward‑deployed researcher and hand over a clean pipeline.

Required profile

• Has shipped post‑trained models into production and can discuss trade‑offs openly.
• Hands‑on depth with Supervised Fine‑Tuning and Reinforcement Learning (DPO, PPO, etc.).
• Ability to evaluate results honestly and recognize misleading metrics.
• Comfortable owning data pipelines: curation, filtering, labeling, and synthetic generation.
• Proficient in PyTorch or JAX.
• Willing to work closely with customers’ domain experts to translate intuition into reward functions.

Required skills

• PyTorch
• JAX
• Reinforcement Learning
• Supervised Fine‑Tuning
• Model Distillation
• Quantization
• Speculative Decoding

What we offer

• Base salary $275,000–$315,000 per year.
• Meaningful equity and comprehensive benefits.
• Relocation assistance and onsite work in San Francisco.
• Access to cutting‑edge compute infrastructure and a small, high‑impact team.
• Opportunity to shape the future of AI compute stack.