Forward Deployed AI Engineer (Google Workspace, Vertex AI)

Limestone Digital; Brasil; Hybrid; BRL 500,000 - 750,000; Full time

Job summary

Limestone Digital in Brazil seeks a software engineer to own the delivery of AI agents and automations for a US-based financial services client. You will design workflows, decide what runs code vs model, and deploy on Vertex AI Engine or Cloud Run, integrating Google Workspace and Slack.

Strong Python and LLM experience, plus production-grade security practices, are essential. Excellent English communication is required for daily client calls and documentation.

Qualifications

• 5+ years of software engineering experience.
• Production experience with LLM applications and guardrails.
• Excellent written and spoken English communication.
• Strong Google Cloud/Vertex AI experience.
• Experience with OAuth 2.0 and least-privilege access.

Responsibilities

• Own delivery of agents and automations from the first workflow walkthrough to production and support.
• Turn documented client processes into a design: which steps are code, which use a model, where a human approves.
• Build agents in ADK, deploy on Vertex AI Agent Engine or Cloud Run, connect to Google Workspace and Slack.
• Set up evaluation and regression checks so each agent is retested when a model version changes.
• Run a short daily or near-daily sync with the client contact during her morning hours.
• Write short runbooks and walk client users through how each solution works; keep monthly work plans and status current for the architect and account lead.
• Follow the client confidentiality policy: approved tools only, no AI note takers, no client data outside approved locations.

Skills

Python

LLM workflows

Google Cloud

Google Workspace APIs

OAuth 2.0 security

Webhooks & event-driven

Tools

Vertex AI Agent Engine

Cloud Run

Apps Script

Job description

Requirements

• 5+ years of software engineering, strong Python
• Production experience with LLM applications: tool calling, structured output, prompt and context design, evaluation, guardrails
• Hands-on Google Cloud. Vertex AI and ADK preferred, solid GCP serverless experience accepted
• Google Workspace APIs in production (Calendar, Gmail, Sheets) and Apps Script
• OAuth 2.0, service accounts, domain-wide delegation, least-privilege access
• Integration engineering: webhooks, event-driven flows, retries, idempotency, logging and alerting
• Builds deterministic workflows first and adds the model where it pays off
• Upper-intermediate or better spoken and written English, comfortable running calls with business users alone
• Careful with sensitive data and strict confidentiality rules

Nice-to-Have

• Vertex AI Agent Engine, Memory Bank, Gen AI evaluation service
• Slack apps (Bolt, Block Kit, slash commands, interactive approvals)
• Calendar scheduling logic: availability search, constraints, time zones, recurring events
• Admin SDK Reports API or other Workspace audit tooling
• MCP servers and clients, A2A
• Claude or Gemini models in production, model routing
• Financial services, private equity or executive assistant workflows
• Experience teaching non-technical users

Responsibilities

• Own delivery of agents and automations from the first workflow walkthrough to production and support
• Turn documented client processes into a design: which steps are code, which use a model, where a human approves
• Build agents in ADK, deploy on Vertex AI Agent Engine or Cloud Run, connect to Google Workspace and Slack
• Set up evaluation and regression checks so each agent is retested when a model version changes
• Run a short daily or near-daily sync with the client contact during her morning hours
• Write short runbooks and walk client users through how each solution worksKeep monthly work plans and status current for the architect and account lead
• Follow the client confidentiality policy: approved tools only, no AI note takers, no client data outside approved locations

About project

The client is a US-based financial services firm that runs fully on Google Workspace and Slack. The project builds AI agents and automations for the firm’s internal operations: scheduling, OKR tracking, email handling, and team requests across the executive office and operations teams. The focus is on reliable, production-grade solutions. Deterministic logic comes first, and LLMs are used where they add real value, backed by evaluation and human oversight.