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

Senior Post-Training ML Engineer, Grounded Reasoning

San Francisco, CA · Full-time · On-site

AI Specialism

Machine Learning

Workplace

On-site

Location

San Francisco, CA

Compensation

$240k – $340k

annual package

Role Dossier & Specification

REF: SENIOR-POST-TRAINI

THE CLIENT & MANDATE

Our client is a category-defining San Francisco AI search, answer, and autonomous research company serving tens of millions of users and thousands of API developers worldwide. Their in-house Post-Training & Reasoning Models group trains custom low-latency grounded models and deep multi-step research agents that interleave chain-of-thought reasoning with live web search, code execution, and strict claim-by-claim citation verification.

They have partnered with Talent247 to hire a Senior Post-Training ML Engineer for Grounded Reasoning onsite in San Francisco, CA. The appointed engineer will design online reinforcement learning (GRPO/PPO) environments and faithfulness reward pipelines that teach models when to search, how to reconcile conflicting live sources, and how to synthesize zero-hallucination answers at blazing speed.

KEY RESPONSIBILITIES

  • Design and scale end-to-end post-training pipelines—combining SFT, DPO, and online multi-turn RL (GRPO/PPO/RLOO)—over dense and Mixture-of-Experts open-weight and custom pretrained backbones.
  • Build live, cached-deterministic web-search and code-interpreter RL environments where models learn adaptive query decomposition, source cross-checking, date-sensitive reasoning, and efficient search-budget stopping.
  • Train high-precision generative reward models and claim-entailment verifiers that penalize unsupported extrapolation, stale facts, sycophancy, and prompt-injected web spam.
  • Distill multi-minute deep-research reasoning trajectories into compact, ultra-fast production models delivering sub-second time-to-first-token at massive consumer QPS.
  • Collaborate closely with neural retrieval engineers to co-train search query planners and answer synthesizers end-to-end.
  • Maintain rigorous live-web, multilingual, financial, coding, and adversarial factuality leaderboards gating every production model rollout.

TECHNICAL ENVIRONMENT & OFFERING

Stack: Python, PyTorch, Verl / OpenRLHF / Megatron-Core / FSDP2, SGLang/vLLM/TensorRT-LLM, Ray, ClickHouse, multi-thousand H100/B200 GPU clusters. Workplace: Onsite in San Francisco, CA. Compensation: $240,000 – $340,000 base salary plus substantial unicorn equity, daily catered meals, and full healthcare & 401(k) benefits.

Core Requirements

ESSENTIAL QUALIFICATIONS

  • 4+ years of hands-on ML engineering or research engineering experience in LLM post-training, RLHF/RLVR, or tool-augmented reasoning models.
  • Proven production experience scaling distributed SFT, DPO, and online RL (GRPO/PPO) runs across 30B–400B+ dense and MoE transformer architectures.
  • Deep familiarity with multi-turn tool-use RL mechanics—asynchronous rollout generation with vLLM/SGLang, environment state caching, off-policy importance sampling, and reward-hacking prevention.
  • Strong instinct for factuality evaluation, citation entailment modeling, and diagnosing subtle hallucination modes under conflicting retrieved evidence.
  • High-velocity execution mindset comfortable shipping model iterations from research ablation to millions-of-users production within weeks.

PREFERRED / BONUS EXPERIENCE

  • Direct experience training DeepResearch-style browsing agents, search-integrated RL models (Search-R1-class paradigms), or code-execution reasoning models.
  • Experience with on-policy distillation or speculative decoding draft-model training for latency-critical consumer serving.
  • Publications at NeurIPS, ICML, ICLR, ACL, or EMNLP in post-training, RLHF, or grounded generation.

Position Telemetry

Location San Francisco, CA
Department Machine Learning
Employment type Full-time
Workplace type On-site
Salary range
$240k – $340k annual package