Principal Research Engineer, Model Training & Post-Training

Inflection AI • Palo Alto, CA

2d • $400,000 - $550,000

About The Position

Inflection’s models are central to our product and platform strategy, and we are looking for a hands-on technical leader to own the model-improvement loop from data and training through evals, post-training, release criteria, and production feedback. This person will sit at the intersection of research, production engineering, and model release, with a mandate to ship models that are measurably better for users. The ideal candidate has led serious model training or post-training work before, can make principled tradeoffs across data, compute, architecture, and quality, around a clear technical roadmap.

Requirements

• Experience leading, or serving as a principal contributor to, large-scale LLM, multimodal, or foundation-model training or post-training programs.
• Deep experience with transformer-based models, hybrid architectures, modern deep-learning frameworks, and distributed training systems.
• Strong practical experience with post-training and alignment methods such as SFT, RLHF, DPO, GRPO, RLAIF, reward modeling, preference optimization, tool-use fine-tuning, or related approaches.
• Experience operating or partnering on large-scale training infrastructure, ideally including GPU clusters at the scale of 1,000+ GPUs.
• Strong systems instincts around throughput, cost, reliability, observability, debugging, checkpointing, reproducibility, and fault tolerance.
• Excellent judgment around data quality, evaluation design, model regressions, release readiness, and production model behavior.
• Ability to balance research ambition with product pragmatism, user impact, and operational discipline.
• Experience leading senior technical teams while continuing to contribute directly to technical decisions and implementation.
• PhD in Computer Science, Machine Learning, Artificial Intelligence, or a related field, or equivalent practical experience.

Responsibilities

• Own the model-improvement roadmap across capability, reliability, emotional intelligence, tool use, safety, latency, cost, and enterprise readiness.
• Lead training and post-training strategy, including supervised fine-tuning, RLHF, DPO, GRPO, RLAIF, reward modeling, preference optimization, tool-use fine-tuning, distillation, synthetic data, and related methods.
• Drive model architecture and optimization decisions across modern transformer-based and hybrid architectures, including both training-time and inference-time performance.
• Lead large-scale training efforts on distributed GPU clusters, including systems operating at the scale of 1,000+ GPUs.
• Define and execute data strategy across data curation, mixture design, deduplication, decontamination, human-in-the-loop pipelines, preference data, evaluation data, synthetic data, and production feedback loops.
• Build and improve evaluation and release-quality systems, including model evals, quality gates, regression detection, release criteria, model-readiness reviews, and post-release monitoring.
• Partner closely with infrastructure and research engineering teams to improve distributed training reliability, checkpointing, fault tolerance, observability, reproducibility, and cost-performance tradeoffs.
• Debug and improve model behavior across the full stack: data, training, post-training, evaluation, infrastructure, product integration, and production feedback.

Benefits

• Diverse medical, dental and vision options
• 401k matching program
• Unlimited paid time off
• Parental leave and flexibility for all parents and caregivers
• Support of country-specific visa needs for international employees living in the Bay Area

What This Job Offers

• Job Type: Full-time
• Career Level: Principal
• Education Level: Ph.D. or professional degree