Research Engineer - Pretraining

Prime Intellect, Inc.; San Francisco (CA); Hybrid; USD 150,000 - 350,000; Full time

Benefits offered by this job

Cash compensation: $150k-$350k incl. 1

Remote or SF office in San Francisco

Job summary

Prime Intellect, Inc. is building the open frontier AI stack and unifying compute, environments, evaluations, secure sandboxes, and deployment into one full-stack system for post-training at frontier scale.

The role focuses on taking open models from architecture experiments to full pretraining runs, coordinating across model and training stacks to ensure successful iterations. You will work on MoE and attention designs, profiling and optimizing training throughput, and building robust,

Qualifications

• Hands-on experience training language models and modifying training code.
• Strong Python skills and PyTorch experience with distributed training debugging.
• Understanding of model architectures and optimization and their effect on convergence.
• Ability to design ablations to answer research questions and explain results.
• Good judgment on when to investigate independently or with a colleague.

Responsibilities

• Build and extend pretraining code in prime-rl and scale experiments to full training runs.
• Evaluate architecture changes (MoE, attention) for quality and cost.
• Profile training jobs and improve end-to-end throughput with hardware-aware execution.
• Make long runs resilient with reliable checkpointing and recovery.
• Investigate training dynamics and numerical stability while preserving convergence.
• Run controlled ablations to guide larger experiments and ensure reproducibility.
• Collaborate with data and evaluation researchers on training mixtures including mid-training contexts.

Skills

LM training

Python + PyTorch

Model architectures

Ablation design

Judgment & initiative

Tools

PyTorch

Job description

Own Your Intelligence

Prime Intellect is building the open superintelligence stack: the infrastructure frontier AI labs build internally, made available to every ambitious AI team.

Our platform, Lab, unifies compute, environments, evaluations, secure sandboxes, high-performance training, and deployment into one full-stack system for post-training at frontier scale - from SFT and RL to tool use, agent workflows, and continuously improving production models. We are building open frontier AI: open-source models trained end to end for long-horizon tasks like autonomous research, and the full-stack platform our own research team uses to build them. The next generation of AI companies, enterprises, and research teams do not just need more GPUs. They need the ability to turn their own workflows, tools, data, and feedback loops into superintelligence they own.

Prime Intellect has raised $150M in total funding from Founders Fund, Radical Ventures, NVIDIA, and exceptional AI, infrastructure, and enterprise operators, including Andrej Karpathy, Dwarkesh Patel, and leaders and founders from Ramp, Perplexity, Harvey, Mercor, Zapier, Datadog, Cognition, OpenAI, Thinking Machines, Together AI, SemiAnalysis, LangChain, Browserbase, Cloudflare, Sierra, Databricks, Airbnb, OpenRouter, Standard Intelligence, Fleet, Core Auto, and more. We are looking for people who want to build at the intersection of frontier research, real infrastructure, and go-to-market for a category that does not fully exist yet.

Role Impact

Take Prime Intellect's next open models from architecture experiments through full pretraining runs, scaling toward 1-3T parameters. You'll work across the model and training stack, making research decisions and fixing the systems problems that determine whether a run succeeds.

Responsibilities

• Build and extend the pretraining code in prime-rl, taking small experiments into full training runs.

• Evaluate model architecture changes, including MoE and attention designs, against model quality and training cost.

• Profile training jobs and improve end-to-end throughput, working with systems engineers on hardware-aware execution.

• Make long runs resilient to failures, with reliable checkpointing and recovery.

• Investigate training dynamics and numerical stability, checking that optimizations preserve convergence.

• Run controlled ablations to guide larger experiments, keeping the setup and results reproducible.

• Work with data and evaluation researchers on training mixtures, including midtraining and long-context experiments.

Requirements

• Hands-on experience training language models, including modifying training code and following the effects through actual runs.

• Strong Python skills and experience with PyTorch or a comparable framework, with enough systems understanding to debug distributed training.

• A practical understanding of model architectures and optimization, including their effect on convergence.

• Ability to design an ablation that answers a research question and explain what its results do and don't show.

• Good judgment about what to investigate yourself and when to bring in a colleague.

Benefits and Perks

• Cash Compensation Range of $150-350k, including equity incentives, aligning your success with the growth and impact of Prime Intellect.

• Flexible work arrangements, with the option to work remotely or in-person at our offices in San Francisco.

• Visa sponsorship and relocation assistance for international candidates.

• Quarterly team off-sites, hackathons, conferences and learning opportunities.

• Opportunity to work with a talented, hard-working and mission-driven team, united by a shared passion for leveraging technology to accelerate science and AI.

If you're excited about the opportunity to build the foundation for the future of decentralized AI and create a platform that empowers developers and researchers to push the boundaries of what's possible, we'd love to hear from you.