Applied AI Scientist (Staff Level)

About the role

We're hiring a senior IC who pairs deep research training with the ability to ship. You'll work on hard reasoning, retrieval, and orchestration problems, and you'll take them all the way into the product. This is research-informed product work: novel thinking is expected, but prototypes are the start, not the deliverable. You'll be a peer to our senior technical leaders and a close partner to product and engineering. We strongly prefer candidates based in the SF Bay Area who are available to work in a hybrid environment.

What you'll do

• Design grounding and context assembly: retrieval pipelines, context selection, query understanding.
• Improve orchestration and reasoning across multi-step workflows: tool use, routing, decomposition, evaluation.
• Build structured representations that help the system reason over accounts, people, and activity.
• Bring research-grade rigor to ambiguous problems, then translate it with engineering into reliable, production-ready capabilities.
• Write real code that holds up in production; you won't be a full-time SWE.
• Raise technical rigor: share patterns, mentor, and accelerate decisions across teams.

Must have

• PhD in CS, ML, NLP, IR, or a related field.
• 7+ years in applied ML, NLP, IR, or ranking (research-track and graduate work can count toward depth), with recent hands-on ownership of LLM and agentic systems.
• Shipped LLM or RAG-style systems into production and personally owned their reliability, latency, and monitoring at scale.
• Built and owned agentic or multi-step orchestration (tool use, routing, planning), not just prototypes or demos.
• Strong coding ability and real comfort in a production codebase.
• Solid experimentation and evaluation rigor: offline evals, A/B testing, causal thinking.
• IC5 signals: set technical direction across teams, owned architecture, and operated as a peer-level influence, not just a strong executor.

Strong plus

• Knowledge graphs, entity resolution, or representation learning.
• Query understanding and intent modeling.
• Search and retrieval depth: lexical, semantic, hybrid, reranking.
• LLM fine-tuning.
• A record of taking research from idea to published result to shipped feature.

You'll do well here if

• You take a nebulous problem, give it structure, build something testable, and iterate.
• You're the glue between data science, ML engineering, and product.
• You explain technical tradeoffs clearly and stay focused on what reaches the customer.