Production AI agents on GCP

I am Rashid Azarang. I build the software layer that lets language models use tools, read business records, run in cloud runtimes, and leave a trace after they act.

The work combines Python, Google ADK, Agent Registry, Vertex AI/Gemini, Cloud Run, CI/CD, observability, IAM, Secret Manager, MCP, and versioned agent artifacts.

Rashid Azarang | rashidazarang.com/work/ai-engineer-gcp-agents | https://github.com/rashidazarang

Current engagement

Conversational agents for 700,000+ students

Production multi-agent conversational systems on Google Cloud for a higher-education institution in Mexico, serving a student population of 700,000+ across admissions, enrollment, scheduling, routing, and student support. Contract engagement.

  • Dominant author of the production agent: prompts, tools, business rules as code, and the appointment and routing logic.
  • Built the evaluation harness end to end — 80+ test modules, 55 golden cases — and calibrated the LLM judge until its verdict could block a release.
  • Evaluation runs as a deploy gate with per-environment depth, on a Terraform-managed GCP estate: Vertex AI, Google ADK, Cloud Run, IAM, Secret Manager, Cloud Monitoring.

What I build

The shape is practical: agents that call tools, touch business data, and can be deployed, tested, versioned, and reviewed.

Multi-agent architecture

Agents built on Google ADK and deployed through Agent Engine, with tools and MCP servers catalogued in Agent Registry. Behavior lives in versioned prompts, business rules as code, and tool contracts — not in a prompt someone edits in a console.

Google ADKAgent EngineAgent RegistryMCPtool calling

Vertex AI and Gemini in production

Model configuration, grounding, and native online evaluators on Vertex AI — and the boundary that matters in production: the model proposes, the system commits. Nothing irreversible happens on model output alone.

Vertex AIGeminionline evaluatorsgrounding

Evaluation engineering

Golden sets as code, an LLM judge calibrated case by case against hand-labeled turns, majority voting so a verdict is stable enough to block a release, hermetic per-case state, and parallel execution once cases are genuinely independent.

golden setsLLM judgecalibrationregression gateshermetic cases

Cloud delivery

Cloud Run services, Docker, a Terraform-managed estate, and CI/CD pipelines whose evaluation gates differ by environment — smoke on integration, full regression on QA — under trunk-based promotion.

Cloud RunDockerTerraformCI/CDAzure DevOps

Identity and enterprise security

IAM as the access model, Secret Manager for credentials, and PII boundaries enforced in telemetry — what the agent may do, what operators may see, and what never leaves the environment.

IAMSecret ManagerPII boundariesaccess control

Observability and cost

Traces and evaluator output exported to Cloud Monitoring, evaluation results as queryable data on BigQuery rather than logs read once, and cost per conversation tracked as a first-class figure.

Cloud MonitoringBigQuerytracingcost per conversation

Role fit

The role asks for an engineer who can move agents from prototype to governed cloud runtime.

Build and deploy agents on GCP

Google ADK, Agent Registry, Vertex AI/Gemini, Cloud Run, Docker, and CI/CD.

Connect agents to business systems

MCP tools, REST APIs, Airtable records, Cotizera workflows, and Dataware-backed operations.

Govern agent actions

Versioned prompts and tools, golden-set regression gates, observability, IAM, Secret Manager, and evidence trails.

Selected work

A few public systems and case studies that show the same pattern in different environments.

MCP gateway

MetaMCP

Gateway work for reducing and exposing MCP tool surfaces.

Repo

MCP server

Airtable MCP

Airtable bases, tables, records, and schema exposed through MCP.

Repo

Python agent

Airtable AI Agent

Python workflows using Airtable tools and business records.

Repo

Business workflow

Cotizera Agents

Quote intake, PDF generation, WhatsApp follow-up, and pipeline updates.

Case

Runtime work

WebHarness

A macOS runtime for web apps with filesystem, SQL, vault, MCP, agents, and packaged distribution.

Site

Evidence layer

Mentu Protocol

Ledgers, hashes, commitments, and evidence for agent actions.

Repo

Data and operations

Agents only become useful when the records underneath them are clean, queryable, and close to the workflow.

Data platform

Dataware

Managed warehouse and MCP access for business data across 50+ integrations.

Site

Dataware

GreenLight

Supabase warehouse with MCP, non-technical access, and issue detection.

Source

Dataware

HWG and TecAssured

Claims, documents, 13 collections, relational warehouse, materialized views, and indexes.

Source

Case study

From Sync Bridge to Data Warehouse

Syncs reduced from 45 to 60 minutes to 13 to 20 minutes. Error rate under 1 percent.

Case

Case study

Building an Enterprise Analytics Platform

MongoDB to PostgreSQL, React dashboard, ETL, 18,000+ dealers, and 50,000+ claims.

Case

Review links

Google sources

Google Cloud

Agent Registry overview

Catalog for agents, tools, MCP servers, and endpoints.

Source

Google Cloud

Register MCP servers

External MCP servers require registration and a manual toolspec.

Source

Google Cloud

Host MCP servers on Cloud Run

Cloud Run supports remote MCP servers over Streamable HTTP.

Source

Google ADK

ADK Agent Registry integration

ADK can fetch MCP toolsets from Agent Registry.

Source