AI Alignment Research Engineer (Post-Training)

Character.AI; California; On-site; USD 225,000 - 400,000; Full time

Benefits offered by this job

Health coverage

401(k) contribution

Paid parental leave

PTO 4 weeks

Doordash stipend

Wellness stipend

Job summary

Character.AI is seeking a Post-Training Researcher on the Post-Training team to advance fine-tuning, alignment, and efficiency of our large language models. You will collaborate across teams to enhance model performance, data pipelines, and training methods, shaping the conversational experiences of millions of users daily.

The role emphasizes production-grade coding, research excellence, and working with GPU workloads to push the boundaries of AI alignment and applicability.

Qualifications

• PhD or equivalent required.
• Ability to write clean production-grade code.
• Experience with GPUs for training, serving, debugging.
• Experience building data pipelines and data infrastructure.
• Strong knowledge of modern ML techniques (reinforcement learning, transformers).
• Proven track record of exceptional research or applied ML projects.

Responsibilities

• Develop alignment algorithms and loss functions to improve data sample efficiency.
• Write data pipelines to process diverse web data into ingestible format for models.
• Identify quality signals to understand model performance in the real world.
• Design sampling algorithms to improve serving efficiency of large generative models.

Skills

PhD or equivalent

Production code

GPUs

Data pipelines

ML techniques (RL/transformers)

Research track record

Education

PhD or equivalent

Job description

Character.AI is seeking a Post-Training Researcher on the Post-Training team to advance fine-tuning, alignment, and efficiency of our large language models. You will collaborate across teams to enhance model performance, data pipelines, and training methods, shaping the conversational experiences of millions of users daily.

The role emphasizes production-grade coding, research excellence, and working with GPU workloads to push the boundaries of AI alignment and applicability.