New opportunity

AI Agent Builder

University of Washington · Seattle, United States of America

About this role

UW Information Technology seeks an AI Agent Builder to design, deploy, and improve AI-powered agents and intelligent automation for the University of Washington’s three campuses, UW Medicine, and research operations. The role reports to a Technology Manager in Infrastructure Service and AI Platforms.

Responsibilities

Design and iterate on AI agents and advanced retrieval-augmented generation pipelines, including embeddings, optimized vector search, context-window management, and hybrid retrieval.
Automate administrative workflows and connect agents to university systems, such as SIS, LMS, HRIS, ERP, and ticketing platforms, through APIs and connectors.
Build research-support tools for literature review, data analysis, grant writing, and knowledge synthesis, including applications using long-context reasoning and grounded research corpora.
Develop tailored tools for faculty and staff, such as drafting assistants, scheduling agents, and data-query tools; implement agent orchestration frameworks.
Monitor performance, usage, and feedback, and implement safety guardrails and evaluation frameworks.

Required qualifications

A bachelor’s degree in Computer Science, Data Science, Software Engineering, or a related field, or equivalent experience; at least 3 years of professional software development experience with strong computer science fundamentals and API integration expertise; AI/ML or LLM-application experience; and a portfolio of deployed agents or automation tools. Candidates need experience building agents, chatbots, or conversational AI with modern LLM frameworks; using LLM APIs and prompt engineering; understanding RAG, embeddings, vector databases, and contextual grounding; and working with REST APIs, AWS, Azure, or GCP, and Docker. Problem-solving and independent and cross-functional collaboration are required. Equivalent education or experience may substitute for minimum qualifications except where legally restricted.

Preferred qualifications

Experience in higher education, research, or the public sector; nebulaONE or comparable orchestration platforms; agent-to-agent coordination and advanced tool use; language-model fine-tuning, evaluation, or alignment; FERPA, HIPAA, and responsible AI; no-code or low-code tools; and open-source AI/ML contributions. MCP integrations are highly desirable; A2A integration familiarity is a plus, not required.

This regular, full-time hybrid position requires at least three days per week in the Seattle U-District office. General hours are Monday–Friday, 8 a.m.–5 p.m., with occasional adjustments and conference attendance or presentations. Annual pay range: $87,624–$142,392; benefits information is available through UW.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

AI agentsIntelligent automationRetrieval-augmented generationEmbeddingsVector searchHybrid retrievalLLMsPrompt engineeringConversational AIAgent orchestrationCrewAInebulaONEREST APIsAPI integrationPineconeWeaviateChromaDBpgvectorAWSAzureGoogle Cloud PlatformDockerOpenAI APIAnthropic APIMCPAgent-to-agent coordinationModel evaluationResponsible AIFERPAHIPAA

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Build and improve grounded AI agents
The posting allocates 25% of the role to agent design and advanced RAG, including embeddings, vector-search optimization, hybrid retrieval, and context-window management. Prepare to explain design and iteration decisions using a deployed agent or automation example.
Automate administrative workflows
Another 25% covers identifying repetitive processes and connecting agents to SIS, LMS, HRIS, ERP, or ticketing systems through APIs and connectors. Prepare to describe how an integration handles system boundaries and failures.
Deliver tailored faculty and staff tools
The role allocates 20% to departmental tools such as drafting assistants, scheduling agents, and data-query tools, alongside agent orchestration. Prepare to explain how user needs shaped a tool and when orchestration added value.
Support research workflows
The posting assigns 15% to literature review, data analysis, grant-writing support, and knowledge synthesis, including long-context reasoning and research-corpus grounding. Prepare to discuss how a tool keeps outputs tied to source material.
Operate and evaluate deployed agents
The remaining 15% covers performance and usage monitoring, user feedback, safety guardrails, and evaluation frameworks. Prepare to describe how evidence from deployment would guide changes to an agent.

Likely challenges to think through

Retrieval quality versus context limits
A likely design challenge is selecting enough relevant material for grounded answers without overloading the context window. Prepare a concrete approach to comparing embedding and hybrid retrieval, inspecting retrieved passages, and measuring answer quality; this is a preparation hypothesis, not a reported UW problem.
Useful automation versus controlled system access
Connecting agents to multiple university systems may create tension between completing workflows and limiting what an agent can read or change. Prepare to discuss scoped API permissions, confirmation before consequential actions, and failure handling. FERPA and HIPAA knowledge is preferred in the posting, not stated as a responsibility for every integration.
Department-specific fit versus maintainability
Tools for faculty, staff, and researchers may need different behaviors, while a shared platform benefits from reusable components. Prepare to explain where you would customize prompts, retrieval, or tools and where you would standardize orchestration and monitoring; this trade-off is inferred from the range of duties.
Agent usefulness versus safe, measurable behavior
Improving user outcomes while adding guardrails can create trade-offs in latency, coverage, and permissible actions. Prepare an evaluation plan that combines task results, safety checks, usage signals, and user feedback; the posting requires these activities but does not prescribe a particular framework.

Practice your answers

Technical practice

Practice: How would you choose embeddings and a vector store for an agent’s RAG pipeline?

What to coverThe role calls for embedding strategies, vector search, and familiarity with vector databases. Suggested preparation: explain how you would compare retrieval quality, indexing and update needs, latency, and operational complexity using representative content.

Practice: When would you use hybrid retrieval, and how would you troubleshoot poor search results?

What to coverThe posting specifically names hybrid retrieval and vector search optimization. Suggested preparation: discuss when lexical and semantic matches complement each other, then describe how you would inspect queries, retrieved passages, and ranking before changing the pipeline.

Practice: How would you manage a limited context window without losing grounding?

What to coverContext-window management and contextual grounding are stated RAG responsibilities. Suggested preparation: compare selecting relevant passages, organizing context, and handling insufficient evidence; explain how you would check whether answers remain supported by retrieved material.

Practice: How would you build and refine a conversational agent using an LLM API?

What to coverThe qualifications call for agent or chatbot development, LLM APIs, and prompt engineering. Suggested preparation: describe prompt and tool boundaries, testing with realistic requests, and how you would diagnose failures before changing a prompt or model.

Practice: How would you connect an agent to an administrative system through a REST API?

What to coverThe role includes automating workflows and integrating with systems such as SIS, LMS, HRIS, ERP, and ticketing platforms. Suggested preparation: walk through a bounded workflow, addressing access control, validation, error handling, and when a person should review an action.

Practice: How would you design a research-support tool that synthesizes a corpus and structured data?

What to coverThe posting names literature review, data analysis, grant-writing support, long-context reasoning, structured data augmentation, and research-corpus grounding. Suggested preparation: explain how you would preserve provenance, distinguish source findings from generated synthesis, and evaluate usefulness without overstating evidence.

Practice: When does a faculty or staff tool need agent orchestration rather than a simpler workflow?

What to coverThe role includes bespoke drafting, scheduling, and data-query tools and calls for orchestration frameworks; MCP integration is highly desirable, while A2A familiarity is a plus. Suggested preparation: compare a single agent with coordinated tools or agents, including failure handling and the added complexity of integrations.

Practice: How would you evaluate and improve an agent after deployment?

What to coverThe posting calls for monitoring performance, usage analytics, user feedback, and evaluation frameworks. Suggested preparation: propose task-specific quality and safety checks, explain how feedback would be investigated, and describe how you would compare iterations before release.

Practice: What guardrails would you consider for agents serving university and health-related operations?

What to coverThe role calls for responsible-AI safeguards, and FERPA and HIPAA knowledge is preferred. Suggested preparation: discuss data minimization, authorized access, grounded responses, and escalation of uncertain or sensitive requests without assuming every tool handles the same data.

Practice: How would you package and operate an agent application on a cloud platform?

What to coverThe qualifications include experience with AWS, Azure, or GCP and Docker. Suggested preparation: outline containerization, configuration and secret handling, deployment checks, and diagnosing an API-dependent service failure without assuming a particular UW infrastructure.

Behavioural practice

Practice: Tell me about a time you shaped a tool with input from nontechnical stakeholders.

What to coverThe role involves tailored faculty and staff tools and daily interaction with subject matter experts and stakeholders. Suggested preparation: use a real situation, your actions to clarify needs and trade-offs, and the resulting change or decision.

Practice: Describe a time you independently solved an ambiguous technical problem.

What to coverThe qualifications call for problem-solving and the ability to work independently. Suggested preparation: explain the situation, how you investigated and tested options, when you sought input, and the result.

Practice: Give an example of improving a deployed tool in response to evidence.

What to coverOngoing agent enhancement, usage monitoring, and user feedback are stated responsibilities. Suggested preparation: describe the initial issue, the evidence and action that guided an iteration, and what changed afterward.

Practice: Tell me about coordinating work across teams with different workflow needs.

What to coverThe role spans administrative processes across departments and requires cross-functional collaboration. Suggested preparation: give a real situation, explain how you reconciled requirements or constraints, and state the outcome.

Practice: Describe a time you communicated a technical limitation or risk to a mixed audience.

What to coverThe posting describes interactions with stakeholders at all levels and occasional conference presentations. Suggested preparation: use a real situation, explain how you adapted the message and proposed a path forward, and describe the result.

Sharpen your knowledge

Topics to prepare

RAG design and grounding
The role calls for embeddings, optimized vector search, hybrid retrieval, context-window management, and grounded answers. Practice suggestion: sketch a retrieval pipeline and explain when hybrid search or a different context-selection strategy would improve a poor answer.
Agent orchestration and tool use
The posting calls for AI agents and orchestration frameworks such as CrewAI or equivalent platforms; nebulaONE or a similar environment is preferred. Practice suggestion: diagram how a drafting agent would choose tools, handle a failed tool call, and decide when to stop.
University-system integration
Agents may connect through APIs and connectors to SIS, LMS, HRIS, ERP, and ticketing systems. Practice suggestion: trace a hypothetical ticketing workflow, identifying what data the agent reads, what action it takes, and where human review would be prudent.
Research support and long-context reasoning
The role includes literature review, data analysis, grant-writing support, and synthesis grounded in research corpora. Practice suggestion: outline how to keep a literature summary tied to its source passages when the corpus exceeds the model’s context window.
Evaluation, guardrails, and privacy
Monitoring, evaluation, and safety mechanisms are responsibilities; FERPA, HIPAA, and responsible AI knowledge are preferred. Practice suggestion: define a small test set for a faculty-facing assistant, including groundedness, inappropriate disclosure, and failure-handling cases.

Skills to revise

Implement a retrieval pipeline
The posting requires familiarity with vector databases and embedding-based retrieval. Practice suggestion: use a small public corpus and one listed option—Pinecone, Weaviate, ChromaDB, or pgvector—to compare vector-only and hybrid retrieval on the same questions.
Build and test an LLM-powered agent
Hands-on LLM API use, prompt engineering, and experience building agents or chatbots are required. Practice suggestion: build a small drafting or scheduling agent, test ambiguous requests and tool failures, and record the prompt or tool changes that improve its behavior.
Connect an automation workflow through REST
REST API integration is required for the role’s administrative automation work; MCP integrations are highly desirable. Practice suggestion: connect an agent to a mock ticketing API, test read and update actions, and document error handling. Treat MCP as an optional extension.
Package and explain a deployed example
The posting calls for Docker and cloud-platform experience, plus evidence of deployed agents or automation tools. Practice suggestion: containerize a small agent, run it in an AWS, Azure, or GCP environment, and prepare a concise architecture and deployment walkthrough.
Measure and improve agent behavior
The role monitors performance, usage, and feedback and implements evaluation frameworks and guardrails. Practice suggestion: run a repeatable set of grounded-answer and unsafe-request tests, record results, make one change, and compare outcomes.

Your preparation checklist

Map the required qualifications to evidence
Prepare a one-page inventory of examples covering professional software development, AI/ML or LLM applications, LLM APIs and prompting, REST integrations, cloud platforms, Docker, and independent and cross-functional work. The posting asks for evidence of these qualifications.
Prepare a deployed-agent walkthrough
Select a deployed agent or automation tool from your portfolio. Be ready to show its purpose, architecture, your contribution, deployment approach, results, and an iteration driven by feedback; the posting explicitly requests evidence of deployed work.
Practice a RAG design exercise
Sketch an agent that answers questions from a grounded corpus. Explain embedding choices, vector-database use, hybrid retrieval, context-window management, and how you would test answer grounding—areas named in the role.
Diagram an administrative integration
Using a hypothetical SIS, LMS, HRIS, ERP, or ticketing workflow, draw the API and connector boundaries, agent actions, failure handling, and human review points. These systems are examples in the posting, not a stated implementation assignment.
Outline a research or staff-facing tool
Prepare a brief design for a literature-review, knowledge-synthesis, drafting, scheduling, or data-query assistant. Show how it would use grounded material or structured data and how you would clarify needs with faculty, staff, or researchers.
Explain orchestration and tool-use choices
Prepare a short comparison of an orchestration framework you know with CrewAI or a comparable platform. If relevant to your experience, show an MCP integration; treat A2A familiarity as a plus, not a requirement.
Bring an evaluation and safety plan
Draft example tests, monitoring signals, and a feedback loop for a deployed agent. Include guardrails and privacy questions appropriate to university and UW Medicine settings; FERPA and HIPAA knowledge is preferred in the posting.
Prepare questions for the employer
Ask which agent and automation use cases are priorities, what enterprise-system access and research corpora are available, how the team evaluates deployed agents, and how faculty and staff feedback shapes iterations. Confirm expectations for the stated Seattle U-District office schedule.

Availability last checked: Nov 4, 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in United States of America. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…