Staff Reasoning Model & RLHF Engineer – Post-Training & Test-Time Compute

Opportunity Overview

Location: 100% Remote (United States)
Employment Type: Full-Time Direct-Hire via Accel DNA
Compensation Range: $210,000 – $250,000 Base + Equity
Hiring Agency: Accel DNA – Premier Frontier Data & AI Talent Partner

Advance state-of-the-art reasoning architectures, developing post-training reinforcement learning loops, Monte Carlo tree search (MCTS) test-time compute, and preference alignment pipelines for frontier LLMs.

Key Responsibilities:

• Architect and scale post-training RL pipelines utilizing DPO, PPO, and Kahneman-Tversky Optimization (KTO) for multi-step reasoning models.
• Implement test-time compute scaling mechanisms, incorporating search-based inference, self-critique loops, and process reward models (PRMs).
• Fine-tune dense and mixture-of-experts (MoE) foundation models for deep mathematical reasoning, multi-turn code synthesis, and analytical tasks.
• Collaborate with AI safety teams to evaluate reward hacking, overthinking behaviors, and reasoning alignment drift.

What You Bring:

• 6+ years of machine learning experience with 2+ years specialized in RLHF, preference fine-tuning, and LLM post-training.
• Proficiency in PyTorch, DeepSpeed, Megatron-LM, vLLM, and distributed GPU training orchestration.
• Demonstrated track record training or adapting open-weights reasoning and code models (e.g., DeepSeek-R1, Qwen-2.5, Llama).
• MS or PhD in Computer Science, Machine Learning, or related quantitative discipline.

How to Apply:

Interested in exploring this opportunity? Submit your profile or portfolio directly to contact@accel-dna.com or apply through https://accel-dna.com/contact-us/.