Member of Technical Staff, Post Training

Profound•San Francisco, CA

12d•$230,000 - $325,000•Onsite

About The Position

As a Post-Training Engineer at Profound, you will turn the company’s proprietary answer-engine data into models that perform measurably better on tasks customers rely on, including analyzing AI visibility, attributing citations, classifying sentiment, verifying claims, using tools, and running long-horizon agentic workflows. Success is measured by whether customers get the right answer, rather than by benchmarks alone. You will own the full model-development loop, from dataset creation and reward design through training and deployment. You will assess model performance with real customers and use production outcomes to guide subsequent training cycles. The role involves collaborating with teams building AI Marketer and analytics products, as well as the Evals team, and taking successful experiments from prototype to production. This is an on-site role based in Profound’s NYC or SF office. The expected base salary range is $230,000 to $325,000. Total compensation includes base salary, equity, and benefits and perks; final compensation depends on factors such as skills, experience, qualifications, and location.

Requirements

Exceptional software engineering skills in Python and PyTorch.
Hands-on post-training experience, such as SFT, RLHF, DPO, GRPO, or similar reinforcement-learning methods.
Experience with distributed training, such as FSDP, DeepSpeed, or Ray, and Kubernetes or Slurm.
Strong evaluation instincts and the ability to recognize when a metric is misleading.
A track record of owning work end to end, from early experiment to shipped model.
Strong opinions about model behavior and staying current with research.
Comfort operating in a fast-paced startup environment with rapid iteration cycles.

Nice To Haves

Experience with agentic or tool-use training, synthetic data, or reward model design.
Experience with retrieval or search-grounded models.
Publications at venues such as NeurIPS, ICML, ICLR, MLSys, ACL, or EMNLP.
JAX experience.

Responsibilities

Turn Profound’s answer-engine data, including prompts, responses, citations, and query fanouts, into high-quality training datasets.
Build and iterate on supervised fine-tuning, preference optimization, and reinforcement-learning pipelines.
Design reward signals that reflect customer value, including reward models, LLM judges, and verifiable rewards.
Train models for tool use and long-horizon agentic workflows in collaboration with the teams building AI Marketer and analytics products.
Work with the Evals team to define what “better” means for each model and evaluate every training run against those standards.
Prototype methods on Profound’s compute and data infrastructure and take successful methods to production.
Ship trained models, own their performance in production, and use regressions and customer behavior to inform the next training cycle.