Job Description & Details
This is a hands-on, 12-month contract building production-grade AI agents using Google's Agent Development Kit. You will not just be hacking together weekend prototypes; you are expected to architect reliable, scalable multi-agent systems that tie directly into heavy-duty enterprise data stores like Snowflake and Salesforce.
What You'll Actually Be Doing
Your day-to-day revolves around agent orchestration, managing state, handling tool use, and setting up sub-agent delegations. You will spend a lot of time deploying these workloads on Google Cloud, wiring up guardrails for observability, and wrestling with latency and cost across model backends. Expect to collaborate closely with data engineering to make sure the agents have clean, well-governed data pipelines to pull from without breaking internal APIs.
The Core Tech Stack
You need rock-solid Python skills and real-world experience shipping LLM applications to production, specifically leveraging frameworks like Google ADK, LangGraph, or AutoGen. A strong grasp of GCP services like Vertex AI and Cloud Run is non-negotiable since that is where these agents will live. Experience hooking into Snowflake or Salesforce will make you instantly effective here, especially when dealing with complex data models and tool-calling schemas.
Interview Expectations
Expect the hiring team to grill you on agent failure modes, specifically how you handle state management and infinite loops during multi-agent delegation. They want to see how you reason about latency and cost trade-offs when calling expensive frontier models versus smaller fine-tuned variants. Be ready to whiteboard a complete architecture for an agentic workflow that needs to securely query a database via function calling while maintaining strict guardrails.
Application Advice
Skip the generic AI fluff on your resume and highlight concrete instances where you moved an LLM or agent system from a Jupyter notebook into a live production environment. Ensure keywords like Google ADK, Vertex AI, Python, agent orchestration, and Snowflake are front and center so you easily sail past the initial ATS filters.