Research Engineer (Pre-training & Post-training)

Waveforms; San Francisco (CA); On-site; USD 200,000 - 250,000; Full time

Job summary

Join a forward-thinking company at the forefront of audio intelligence! As a Research Engineer, you'll play a pivotal role in the lifecycle of AI models, focusing on pre-training and post-training processes. Your expertise will help build and optimize large-scale data pipelines, ensuring the efficient handling of multimodal datasets. You'll employ cutting-edge techniques like reinforcement learning and generative modeling to refine models and enhance their performance. This is an exciting opportunity to push the boundaries of AI, contributing to innovative solutions that transform human-AI interactions. If you are passionate about audio intelligence and eager to make an impact, this role is for you!

Qualifications

• Experience in training and optimizing large language models, including pre-training and fine-tuning.
• Strong background in distributed systems and data pipeline management for multimodal datasets.

Responsibilities

• Lead pre-training and fine-tuning of large-scale language models for audio and text.
• Develop scalable data pipelines and optimize model performance using advanced techniques.

Skills

Training large language models (LLMs)

Compute efficiency

Distributed systems

Data pipeline management

Reinforcement learning from human feedback (RLHF)

Generative Adversarial Networks (GANs)

Diffusion models

Python

PyTorch

Tools

AWS

GCP

Azure

Fully Sharded Data Parallel

Job description

Job title: Research Engineer (Pre-training & Post-training) / Member of Technical Staff

Who We Are
WaveForms AI is an Audio Large Language Models (LLMs) company building the future of audio intelligence through advanced research and products. Our models will transform human-AI interactions making them more natural, engaging and immersive.

Role overview: The Research Engineer – Pre-training & Post-training role integrates responsibilities across all phases of the AI model lifecycle, including pre-training, post-training, and data preparation. This position involves building and optimizing large-scale data pipelines, handling multimodal datasets (audio and text), conducting pre-training with a focus on compute efficiency and scalability, and refining models with cutting-edge techniques like supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF) and generative modeling. The ideal candidate will leverage advanced methods, including GANs and diffusion models, to push the boundaries of multimodal AI systems focused on audio and text.

Key Responsibilities

• Lead the pre-training and fine-tuning of large-scale language models (LLMs), maximizing compute efficiency and scaling infrastructure.

• Optimize model performance using advanced techniques, including RLHF, reward modeling (RM), instruction-tuning, distillation, GANs, and diffusion models.

• Develop robust evaluation pipelines to monitor, refine, and improve model performance throughout training phases.

• Build and optimize scalable, distributed data pipelines to support multimodal (audio + text) AI training.

• Handle and process massive datasets (PiB scale) for pre-training and post-training, ensuring efficient preparation, annotation, and data flow.

• Collaborate with research and engineering teams to ensure seamless integration of data preparation and training workflows for multimodal systems.

Required Skills & Qualifications

• Proven experience in training large language models (LLMs), including pre-training, fine-tuning, and post-training optimization.

• Strong background in distributed systems, compute efficiency, and scaling model training infrastructure.

• Expertise in designing and managing large-scale, distributed data pipelines for multimodal datasets, particularly audio + text.

• Proficiency in advanced techniques such as RLHF, instruction-tuning, reward modeling, distillation, GANs, and diffusion models.

• Proficiency in Python, PyTorch, and distributed frameworks (e.g., Fully Sharded Data Parallel)

• Familiarity with cloud platforms like AWS, GCP, or Azure for managing distributed environments.

• Knowledge of multimodal AI systems combining audio and text for training and evaluation.