Model Job Description for AI Engineer, Search & Knowledge Systems

Role purpose

Build and improve a search or retrieval-augmented knowledge feature that returns useful information from approved sources to authorized users. Make the source-to-answer path inspectable: what information entered the system, how it was represented and retrieved, which identity could access it, whether a cited passage supports the result, and how the feature behaves under ordinary operating constraints. The role works with data, domain, product, application, platform and security colleagues where those functions exist. AbbVie source-preparation role; Glean API role; Pi Security search role.

Core responsibilities

• Clarify the user's information task, source owners, authorized audiences, freshness needs, acceptable failure behavior and measurable quality criteria before choosing a retrieval approach. Capture disagreements and decisions in a short design note. Requirements discovery with domain experts and design trade-off documentation appear in the AbbVie and Accellor roles.
• Prepare source material for retrieval. Inventory document types and metadata; validate extraction, reading order, tables and identifiers; implement or coordinate ingestion, transformation and indexing; record update and failure behavior. Scope graph or entity-resolution work only when the source and task require it. AbbVie; Amazon Quick Suite; ServiceNow search infrastructure.
• Implement and compare retrieval behavior appropriate to the case: lexical, vector, structured-filter, hybrid or ranking steps. Connect the selected method to a user-facing application or service and record how a result maps back to its source. A specific posting names vector plus lexical retrieval and structured attributes; another names hybrid search, reranking and citations. Amazon DynamoDB search; Pi Security.
• Build test cases for relevance, source support and authorized access. Include representative tasks, hard negatives, missing or stale sources, denied identities and insufficient evidence. Record observed failures and retest after changes. Evaluation frameworks and entitlement-aware retrieval are stated in Exa's adjacent search evaluation role and Amazon's retrieval engineering role; LTS names retrieval quality, hallucination, latency, throughput and cost.
• Work with application and platform colleagues on APIs, deployment, monitoring and recovery where the assigned role includes those duties. Explain trade-offs in relevance, latency, cost, reliability and security to the accountable decision maker. Glean; LTS; Accellor.

Expected work products

• A short information-task and source-access brief naming users, sources, source owners, refresh needs and success criteria.
• A source inventory and ingestion/indexing design, followed by working queries, a pipeline, index, service or search/RAG feature as assigned.
• A test set and results record showing relevant hits, unsupported or missing answers, authorized and denied access, and known limitations.
• A documented integration contract, design decision or handoff when another team owns the API, platform, source system or release.
• A monitoring, incident or improvement record when the role includes production operations.

These are adaptable examples, not a claim that one posting requires all five. The observed U.S. advertisements separately name pipelines and services, search features, APIs, evaluation artifacts and design documents. Amazon Quick Suite; Glean; Exa; Accellor.

Capabilities and observable behavior

Technical methods

• Explain and test ingestion quality, metadata, document structure, index updates and provenance.
• Compare lexical, vector and hybrid retrieval with a task-specific relevance set; understand when structured filters, reranking, graph relationships or citations add value.
• Build or integrate a service/API with a reproducible query-to-source path.
• Test identity and document-level access, grounding, failure handling and operating limits without treating a plausible answer as a pass.

Working with people

• Ask domain experts to clarify ambiguous terms and source authority, then record the agreed interpretation. AbbVie.
• Present a design choice and its measurable trade-offs to technical and nontechnical partners. Lynx Analytics.
• Give product, backend, platform and security colleagues a specific finding, owner and next action. Pi Security.
• Review or mentor other engineers when that responsibility is part of the actual level. Glean search-quality role.

Tool and experience profile

Choose tools from the organization's approved systems: document stores and connectors; parsing/transformation tools; search indexes, vector or graph stores where warranted; embedding and ranking services; API and application frameworks; identity/ACL controls; evaluation harnesses; and monitoring. Product names in a particular vacancy do not define a universal stack. The LTS platform posting, for example, puts named vector databases and orchestration frameworks in a preferred section.

Set experience requirements against the actual responsibility and support model. A contributor may implement a bounded source, retrieval or evaluation component under an established architecture. A more senior owner may set architecture boundaries, API standards or index lifecycle decisions; those rights must be expressly assigned. Employer examples range from a ServiceNow infrastructure role with 1+ years stated to an Accellor principal role with 10–12 years stated. Do not combine their thresholds into a single requirement.

Interfaces, decisions and escalation

The role identifies and documents options, implements within the agreed design and reports test evidence. The source owner decides which material is authoritative and may be used. The organization's designated identity/security owner approves changes to access rules. The accountable product or service owner accepts release scope and operational risk. A platform/API owner approves interface or deployment changes where the organization assigns that control. These are local role assignments to confirm, not authority inferred from this model. A Glean role explicitly owns API versioning standards, while an Amazon search role helps decide index lifecycle behavior; neither establishes the other organization's approval chain.

Escalate immediately when a test returns a document to a denied identity, a citation does not support a consequential answer, a source's use or freshness is uncertain, or production quality falls outside locally approved limits. Record the failing identity/query/source, impact, containment and the owner who will decide the next action. Resume only through the organization's incident and release process.

Working rhythm

Daily work may include source/pipeline checks, implementation, evaluation and monitoring if assigned. Weekly work may include review of failures, feedback with domain or product partners and design decisions. Monthly review may revisit source ownership, drift, cost and access tests. Event-driven work covers source changes, release regression, incident response, index rebuild and authorization changes. These are adaptable operating prompts, not observed universal frequencies: most reviewed advertisements did not state a task cadence. Specific examples include Amazon's daily dashboard review, Amazon's weekly customer-feedback sync in a solution-architect role, NetSpeek's per-release evaluation, and Smartsheet's weekly production-support rotation.

Quick reference