Senior Staff Applied AI Engineer - Context Retrieval

Навыки

• Data Lake
• Отладка и поиск ошибок
• Elasticsearch
• Дообучение моделей
• GMV / MRR
• LLM
• MLOps

О компании и продукте

• Databricks agents are only as good as the context they can retrieve. Whether an agent is answering a question about last quarter's revenue, debugging a failing job, generating SQL against a 10,000-table lakehouse, or summarizing a Wiki page, its quality is bounded by what it can find — and how well it understands what it finds.
• We are hiring a Senior Staff Applied AI Engineer to own context retrieval for Databricks agents across SaaS providers . This is a zero-to-one role with two deeply connected charters:
• Build the retrieval stack — query understanding, content understanding, ranking, retrieval, and evaluation — across the Enterprise SaaS data stored across multiple systems.
• Build the search subagents that sit on top of that stack and reason about what context is needed , how to retrieve it , and whether the right thing actually came back — closing the loop between an agent's intent and the substrate that serves it.

Задачи

• Build the full retrieval stack from scratch
• Own the end-to-end system: query understanding, content understanding and indexing, hybrid retrieval, ranking, and evaluation
• Make the architectural calls that will define how Databricks agents access context for years to come
• Retrieve across heterogeneous data — structured and unstructured
• Index and rank across structured assets (tables, columns, SQL queries, dashboards, code, notebooks, jobs) and unstructured content (docs, wikis, tickets, chat, images, video, audio)
• Each modality has its own signals — design retrieval that exploits them rather than flattens them
• Connect to the SaaS surface area customers actually use
• Build connectors and retrieval adapters for the systems where enterprise knowledge lives
• Treat each retrieval source with its own freshness, permissions, and ranking signals
• Optimize for two consumers at once
• Retrieval must serve both LLMs (grounded, token-efficient, hallucination-resistant context) and humans (intuitive, explainable discovery)
• These are different objectives and require different signals — own both
• Crack query understanding for agents
• Agent queries don't look like web queries
• Build query rewriting, decomposition, intent classification, and entity resolution tuned for multi-turn agentic workflows
• Crack content understanding at scale
• Build the pipelines that extract structure, entities, embeddings, summaries, and metadata from every supported asset type — and keep them fresh as customer data evolves
• Build search subagents that reason about retrieval
• Design the agentic layer that decides what context is needed , which sources to query , how to decompose and route the search , and — critically — whether the retrieved content is actually sufficient to answer the question
• These subagents will plan multi-hop searches, issue follow-up queries when results are weak, ground claims against retrieved evidence, and hand back high-confidence context (or signal failure) to upstream agents
• This is where IR meets agentic reasoning
• Build the evaluation flywheel for both retrieval and subagents
• Stand up offline evals (nDCG, MRR, Recall@K, Precision@K), LLM-as-judge harnesses, human-in-the-loop labeling, and online experimentation
• Extend evaluation beyond ranking metrics to measure subagent decision quality — did it ask the right follow-up? , did it correctly recognize when retrieval failed? , did it ground its answer in the right evidence?
• Quality you can't measure is quality you can't ship

Требования

• 10+ years of software engineering experience, with significant time spent building production retrieval, search, or RAG systems at scale
• Deep Information Retrieval (IR) expertise : lexical retrieval (BM25, Lucene/Elasticsearch/OpenSearch), dense retrieval (embeddings, ANN indexes — FAISS, ScaNN, HNSW), hybrid retrieval, and learning-to-rank
• Hands-on experience with modern LLM-era retrieval : RAG architectures, query rewriting, re-ranking with cross-encoders, long-context strategies, and grounding techniques that reduce hallucination
• Experience designing agentic systems on top of retrieval — search planners, multi-hop / iterative retrieval, self-reflection and sufficiency checks, tool-using agents that decide what to fetch and verify what came back
• Strong grasp of relevance evaluation : nDCG, MRR, Precision@K, Recall@K
• offline/online experimentation
• LLM-as-judge frameworks
• building human labeling pipelines
• Experience working across structured and unstructured data — you've indexed and ranked over tables, code, and documents in the same system, and have opinions about how to do it well
• Track record of building 0→1 : you've stood up a retrieval system from an empty repo, made the foundational architectural decisions, and grown it into something that customers depend on
• Demonstrated ability to operate as a technical leader : setting direction across teams, mentoring senior engineers, and influencing roadmap with research, product, and platform partners

Будет плюсом

• Experience building retrieval over enterprise SaaS sources (permissions, freshness, multi-tenancy, ACL-aware indexing)
• Background in agentic systems, tool use, or multi-turn retrieval for LLM agents
• Contributions to open-source IR/search projects, or publications at SIGIR, KDD, WWW, EMNLP, or similar venues
• Experience training or fine-tuning embedding models, rerankers, or query understanding models

Условия

• At Databricks, we strive to provide comprehensive benefits and perks that meet the needs of all of our employees

Паспорт вакансии

История публикации

Откуда что взялось

Отмечено то, что вывели мы. Без пометки — значение назвал работодатель.

Почему на этом месте в выдаче

Порядок выдачи объявлен контрактом: свежесть решает между днями, полнота и зарплата — внутри дня.