Skip to content

Cloud Mastery Training Introduction

Overview

Build and deploy SokoAI — a real, production-grade AI-native commerce platform — in a single day using Google Cloud. Three sessions, one complete system: from zero to a deployed app with a conversational agent and ML-powered recommendations.

Detailed Session Breakdown

Session 1 — Serverless & Secure Foundation

Zero to a live, deployed application: Next.js storefront + NestJS API on Cloud Run, backed by Cloud SQL.

Cloud Run Deployment

  • Clone the SokoAI starter repo (pre-seeded GitHub repository provided)
  • Containerise with Docker — walkthrough of the Dockerfile
  • Deploy Next.js frontend and NestJS API to Cloud Run as separate services
  • Configure environment variables, secrets, and Cloud Run service settings

Keyless Authentication

  • Why service account keys are dangerous — the "keys under the mat" problem
  • Workload Identity Federation — concept and hands-on configuration
  • Connect NestJS API to Cloud SQL via Cloud SQL Auth Proxy — no stored credentials
  • Test the connection: NestJS API returns SokoAI product data from Cloud SQL

Session 2 — The Agent Brain

Build the SokoAI multi-agent system from scratch in Vertex AI Agent Builder and wire it into the frontend.

Agent Architecture

  • Vertex AI Agent Builder overview — agents, tools, data stores, playbooks
  • One orchestrator → three specialist sub-agents
  • Persona, tone, and scope: Kenyan English, grounded only in SokoAI data

Building the Three Sub-Agents

  • Soko Shop Agent — products + inventory Tools: search_products, add_to_cart, check_delivery_eta
  • Soko Parts Agent — vehicle-to-battery compatibility matrix Tools: match_part, check_stock, find_nearest_location
  • Soko Wealth Agent — financial products + rule engine Tools: get_products, build_recommendation, calculate_projection

Frontend Integration

  • Call the Vertex AI Agent API from Next.js (server action or API route)
  • Stream agent responses into the chat UI
  • Live test: send a message and watch SokoAI respond from your own agent

Session 3 — Data Intelligence

Every user action becomes a training signal. This session shows how an AI-native app gets smarter from its own usage.

The Data Pipeline

  • Cloud Functions triggered by Cloud SQL events → BigQuery
  • Event schema:
  • session_id, user_id, query_text, module
  • product_viewed, cart_action, order_status
  • Walk through the pre-built pipeline — observe, don't configure from scratch

BigQuery ML Models

  • K-Means clustering: segment users by behaviour (high-value, price-sensitive, parts-focused, wealth-seeking)
  • Classification model: predict browse-to-buy conversion probability
  • All in SQL — no Python, no ML framework

Closing the Loop

  • BQML outputs feed back into agent recommendations (e.g. high-value segment → premium MMF products first)
  • Looker Studio dashboard: agent activity, user segments, and conversion predictions on one screen

Grand Finale

The full SokoAI system live — deployed app, conversational agent, ML-powered recommendations. The room built this today.

What to Bring

  • Laptop with Docker installed (Cloud Shell fallback available)
  • A Google Cloud account — project IDs and credentials provided on the day
  • A browser with access to Google Cloud Console
  • Curiosity and an appetite to build something real
← Previous: Home
Introduction
Next: Automate Securely →