Senior AI Engineer (Search / Retrieval)

As a Senior Software Engineer on our Enterprise Retrieval team, you’ll help build the retrieval layer that powers enterprise AI agents at Workato. Your work will let an agent answer “what’s the status of the Acme renewal?” by stitching together a Salesforce opportunity, a call summary, the latest Zendesk ticket, a Jira blocker, and a SharePoint contract — all in one ranked, permission-aware response.

This is the kind of problem where classical Information Retrieval, dense vector retrieval, knowledge graphs, and LLM-driven reasoning all collide. You’ll work across heterogeneous content (docs, tickets, tasks, CRM records, call transcripts, chat threads), heterogeneous permissions (every source has its own ACL model), and very real freshness constraints (yesterday’s answer is often wrong).

It’s a hands-on Senior IC role for someone who wants to go deep on retrieval quality and see their work directly shape how thousands of enterprises put AI agents to work.

In this role, you will also be responsible to:

• Build a unified retrieval layer across enterprise systems — Google Drive, SharePoint, Confluence, Jira, Asana, Zendesk, Freshdesk, Salesforce, Notion, and more — exposing a clean, agent-friendly interface.

• Design hybrid retrieval pipelines that combine lexical (BM25), dense vector, and structured (SQL/graph) retrieval, with smart re-ranking tuned for cross-source results.

• Engineer ingestion and freshness pipelines that incrementally sync millions of documents, tickets, tasks, and CRM records with low end-to-end latency and predictable cost.

• Own permission-aware retrieval (ACL preservation) — make sure the engine never returns a document a user (or their agent) isn’t entitled to see, mirroring source-system permissions exactly.

• Build query understanding for agents — intent parsing, entity linking across systems (a “customer” in Salesforce is the same as in Zendesk), and LLM-assisted query rewriting and decomposition.

• Design chunking and embedding strategies tailored to each content type — long docs, short tickets, threaded conversations, structured records, call transcripts.

• Build evaluation and experimentation harnesses (NDCG, MRR, recall@k, faithfulness, citation accuracy) for both retrieval and end-to-end agent answers.

• Ship production-grade, observable systems with strong SLOs on latency, freshness, recall, and cost — and the dashboards/tracing to prove it.

• Mentor teammates and raise the bar on retrieval architecture, evaluation rigor, and engineering craft.

• 3-5 years building production search, retrieval, knowledge-base, or recommendation systems.

• Strong proficiency in at least one modern backend language — Python, Go, Java, or similar.

• Hands-on experience with search engines such as OpenSearch, Elasticsearch, Solr, or Vespa, including index design and analyzers.

• Solid grounding in IR fundamentals: TF-IDF, BM25, learning-to-rank, query parsing, and relevance evaluation.

• Working experience with vector search and embeddings — FAISS, pgvector, Pinecone, Weaviate, Qdrant, Milvus, or native Elasticsearch/OpenSearch kNN.

• Experience designing or contributing to RAG pipelines and semantic search systems in production.

• Familiarity with modern NLP/LLM tooling: transformer embeddings, cross-encoder re-rankers, prompt engineering, and frameworks like LangChain, LlamaIndex, or Haystack.

• Comfortable building integrations against SaaS APIs (REST/GraphQL/webhooks), handling OAuth, rate limits, pagination, and incremental sync.

• Solid intuition for ACL/permission models in enterprise systems (Drive sharing, SharePoint groups, Jira project roles, Salesforce sharing rules, etc.) and how to preserve them in a retrieval layer.

• Strong SQL skills, comfort with NoSQL/document stores, and experience with large-scale distributed systems.

• Familiarity with cloud platforms (AWS, GCP, or Azure), containerization, and CI/CD.

• Clear communicator who can explain technical trade-offs to engineers, PMs, and executives alike.

• Collaborative — you partner naturally with ML, product, security, and platform teams.

• Quality-obsessed and detail-oriented, with an instinct for measurable outcomes over vibes.

• Self-directed; comfortable taking an ambiguous problem from zero to shipped.

• Genuinely curious about the hard, interesting problems hiding inside enterprise retrieval — heterogeneity, permissions, freshness, and trust.

• Experience with knowledge graphs, entity resolution, or cross-source identity linking.

• Experience tuning or fine-tuning embedding models (sentence-transformers, BGE, E5, etc.) for domain-specific retrieval.

• Exposure to agentic AI patterns — tool use, function calling, MCP, or multi-step retrieval planning.

• Experience with streaming/real-time ingestion (Kafka, Flink, Spark) and cost optimization at scale.

• Background in enterprise search, e-discovery, observability, or DLP — anywhere you’ve had to handle messy multi-source content with strict access controls.

• Open-source contributions, published research, or writing on retrieval, IR, or applied ML

(REQ ID: 2457)