Skip to main content
TALENT247 // SPECIALIST AI EXECUTIVE SEARCH & TECHNICAL RECRUITMENT
ML · Machine Learning

Senior Machine Learning Engineer, Long-Context Domain Post-Training

San Francisco, CA · Full-time · Hybrid

AI Specialism

Machine Learning

Workplace

Hybrid

Location

San Francisco, CA

Compensation

$225k – $315k

annual package

Role Dossier & Specification

REF: SENIOR-MACHINE-LEA

THE CLIENT & MANDATE

Our client is the category-defining domain AI platform serving the world's top AmLaw 100 firms, global investment banks, and Fortune 500 enterprises from headquarters in San Francisco. Unlike thin prompt wrappers, their core technical moat includes an internal Models & Post-Training research team that trains custom long-context reasoning models specifically optimized for multi-document legal, tax, and regulatory logic.

They have partnered with Talent247 to hire a Senior Machine Learning Engineer for Long-Context Domain Post-Training in San Francisco. Collaborating closely with domain experts (attorneys, tax specialists, M&A analysts) and frontier lab partners, the appointed engineer will advance rubric-guided reinforcement learning and 128k–1M+ token post-training recipes over semi-verifiable professional tasks.

KEY RESPONSIBILITIES

  • Design and execute end-to-end supervised fine-tuning (SFT), preference optimization, and online reinforcement learning (GRPO/PPO/RLOO) runs across open-weight and partner frontier model families.
  • Pioneer rubric-based reward modeling and decomposed generative verifiers for semi-verifiable tasks (contract redlining, statutory conflict analysis, multi-exhibit cross-referencing) where simple unit tests do not exist.
  • Optimize long-context (128k–512k+ token) multi-document reasoning performance, eliminating lost-in-the-middle recall drop-offs, subtle citation fabrication, and premature summarization shortcuts.
  • Build synthetic trajectory generation, expert critique-and-revision loops, and hard-negative data mining pipelines seeded by in-house legal and financial specialists.
  • Partner with the agentic runtime team to co-train models directly inside live multi-step search, citation, and document-editing tool loops.
  • Maintain rigorous contamination-free domain benchmarks measuring win rates against unadapted frontier reasoning models.

TECHNICAL ENVIRONMENT & OFFERING

Stack: Python, PyTorch, Verl / OpenRLHF / Megatron-Core / FSDP2, vLLM/SGLang, FlashAttention-3 / Ring Attention, multi-node H100/B200 GPU clusters. Workplace: Hybrid in San Francisco, CA. Compensation: $225,000 – $315,000 base salary plus high-value equity grants, comprehensive healthcare, and 401(k).

Core Requirements

ESSENTIAL QUALIFICATIONS

  • 4+ years of hands-on machine learning engineering or research engineering experience focused on LLM post-training, alignment, or fine-tuning at scale.
  • Proven track record running distributed SFT, DPO, PPO, or GRPO training jobs on 30B+ or MoE transformer models using PyTorch, FSDP, DeepSpeed, or Megatron.
  • Practical expertise in reward model design, rubric/judge calibration, reward-hacking mitigation, and synthetic reasoning trace curation.
  • Deep familiarity with long-context training mechanics (RoPE scaling, context parallelism / Ring Attention, sequence packing, KV-cache memory profiling).
  • Strong experimental rigor—designing clean small-scale ablations that reliably predict large-scale post-training gains.

PREFERRED / BONUS EXPERIENCE

  • Experience training multi-turn tool-using or RAG-integrated reasoning agents via online reinforcement learning.
  • Publications or technical blog posts on RLHF/RLAIF, Constitutional AI, long-context needle-in-a-haystack reasoning, or domain adaptation.
  • Exposure to high-precision professional domains (law, finance, medicine, tax, or scientific literature).

Position Telemetry

Location San Francisco, CA (Hybrid)
Department Machine Learning
Employment type Full-time
Workplace type Hybrid
Salary range
$225k – $315k annual package