AI Engineer- ML/RL - Weights & Biases

What You'll Do:

The AI team is a hands-on applied AI group at Weights & Biases that turns frontier research into teachable workflows. We collaborate with leading enterprises and the OSS community. We are the team that took W&B from a few hundred users to millions of users and one of the most beloved tools in the ML community.

This is a senior applied role at the research-to-production boundary. You will prototype, evaluate, and ship reusable DL/RL workflows for enterprise use on the W&B stack—then document and teach them to our customers and the community. The focus is application, not novel research: rapid prototyping, careful evaluation, and production-grade reference implementations with clear trade-offs.

About the role:

• Understand the state-of-the-art in deep learning / AI and turn the research into practical workflows that can be adopted by our users, the open source community & enterprise customers alike.
• Build in public: Publish engineering artifacts (code, reports, talks) that teach how to reproduce results; engage with OSS and customer engineers.
• Design and ship reference workflows for post-training & agents (SFT/DPO/GRPO/PPO, reward models, online RLHF/RLAIF) with reproducible repos, W&B Reports, and dashboards others can run.
• Own end-to-end demos: data → distributed training (FSDP/ZeRO/DeepSpeed/JAX pjit) → evaluation (lm-eval-harness + agent benches) → serving (vLLM/TensorRT-LLM/Triton/SGLang).
• Partner with lighthouse customers; turn recurring patterns into templates and product feedback.
• Track recent advances (papers, releases, kernels), run focused ablations, and translate wins into production-ready workflows.
• Run growth experiments to track the usage of the Weights & Biases suite of products from the artifacts built.

Who You Are:

• Deep learning: 5+ years training large models in PyTorch or JAX; strong numerics (autograd, initialization, mixed precision).
• RL/RLHF: hands-on with SFT/DPO/GRPO/PPO, reward modeling, preference data pipelines, and online/offline RL for LLMs/agents.
• Inference/serving: production experience with vLLM/TensorRT-LLM/Triton; quantization, speculative decoding, caching.
• Evaluation: built task/agent harnesses with statistically sound metrics (variance, CIs, power) and failure taxonomies.
• Systems: strong Python plus one: CUDA/Triton kernels, custom C++ ops, or high-performance data ingestion.
• Reproducibility: rigorous experiment tracking (sweeps, artifacts, lineage); minimal repros others can run.
• Public signal: 2+ OSS repos/notebooks/talks with adoption (e.g., stars, forks, downloads, conference views).

Preferred: 

• Paper-to-production within weeks at a top lab or applied-AI team (pretrain → post-train → eval → serve).
• Data engines & feedback loops (rater pipelines, synthetic data, active learning).
• Prior customer enablement with external adoption at scale.

Wondering if you’re a good fit? We believe in investing in our people, and value candidates who can bring their own diversified experiences to our teams – even if you aren't a 100% skill or experience match. Here are a few qualities we’ve found compatible with our team. If some of this describes you, we’d love to talk. 

• You love turning cutting-edge AI research into clean, benchmarked, production-ready templates others can use today.
• You’re curious about RL-based post-training and agent evaluation, and maintain your own leaderboards.
• You’re an expert in at least one of: distributed training at scale, RLHF/GRPO systems, or low-latency LLM serving—and you can demonstrate it with code and benchmarks.

The base pay and target total cash for this position range from $182,000 to $242,000. The starting salary will be determined based on job-related knowledge, skills, experience, and market location. We strive for both market alignment and internal equity when determining compensation. In addition to base salary, our total rewards package includes a discretionary bonus, equity awards, and a comprehensive benefits program (all based on eligibility)