AI Engineer

CoreWeave•Sunnyvale, CA

4d•$182,000 - $242,000•Onsite

About The Position

W&B Models is the experiment tracking platform used by the world's leading AI teams, from frontier labs training foundation models to enterprise teams fine-tuning for production. Researchers live in it daily: logging runs, comparing training curves, debugging divergences, building reports, and deciding what to try next. As Senior Product Manager for Models, you'll drive improvements across the core experiment workflow: how researchers design experiments, visualize what's happening inside them, monitor long-running training jobs, and analyze results across hundreds or thousands of runs. You'll do this at a moment when autoresearch and agentic workflows are reshaping what that loop looks like. Your job is to deeply understand how modern ML research actually gets done and make the inner loop of hypothesis → run → analysis → next run dramatically faster and more insightful. That loop looks different for a frontier researcher running thousand-run sweeps and an enterprise team fine-tuning open models against domain evals and cost targets – you'll serve both. This role reports to the Director of Product Management and partners closely with engineering, design, and teams across W&B and CoreWeave.

Requirements

You've lived the ML researcher's workflow. You've trained models yourself, or worked so closely with researchers that their workflow is second nature. You know what it feels like to babysit a training run, debug a loss spike at 2am, or dig through a hundred runs to figure out which ablation mattered.
You have 3+ years of product management experience on technical products for developer or researcher audiences, with a track record of shipping products users love.
You have strong instincts for data visualization and product design. You can reason about what makes a chart, dashboard, or comparison view genuinely useful vs. merely present, and you sweat the details of how information is displayed.
You reach for AI tools instinctively. Claude, Codex, and agentic workflows are how you work: for prototyping, analysis, and exploring ideas. You're excited to build products for people who work the same way.
You're technically fluent. You can read a training script, discuss evaluation methodology, and hold your own in architecture conversations with engineers.
You communicate with clarity and precision, keeping engineering, design, and go-to-market stakeholders aligned without ceremony.

Nice To Haves

Hands-on experience with experiment tracking tools, with opinions of where they all (especially W&B!) fall short
Experience with enterprise ML platforms or fine-tuning services (SageMaker, Vertex, Databricks, Together, Fireworks), and views on what they get right and wrong
Public technical writing, whether a paper, a conference talk, or blog posts discussing ML topics
"AI Twitter" presence or participation.
Experience conducting user research and/or engaging in developer communities

Responsibilities

Advance the researcher's daily workflow. Experiment tracking is where researchers spend their days (and nights). You'll drive the roadmap for how runs are organized, compared, and understood, with a relentless focus on the questions researchers need to answer.
Rethink how training is visualized and monitored. Modern training runs are long, expensive, and failure-prone. You'll shape how W&B surfaces what matters mid-run — loss curves, system metrics, evals, anomalies — so teams catch problems early and understand their models more deeply, across web and mobile.
Make evaluations a first-class part of experiment tracking. Evals are how teams know whether a model is actually getting better, but today they live at arm's length from the training workflow. You'll drive the roadmap for logging, comparing, and analyzing eval results within Models — across runs, across training steps, and at the row level where regressions actually hide.
Make analysis at scale a first-class experience. As experiments grow from dozens of runs to thousands, the hard problems shift from logging to sense-making. You'll define how researchers slice, aggregate, and reason across large experiment histories, including where AI-assisted analysis can do work humans currently do by hand.
Ground every decision in how researchers actually work. You'll spend real time with users, from frontier lab researchers to individual practitioners, and translate what you learn into product decisions the whole team can rally behind.
Ship with quality and speed. You'll own execution from problem definition through launch and iteration, using usage data and customer feedback to guide each stage. W&B's users are discerning; polish and performance are features.