Work / AI · Agents / APAC GenAI Academy

No. 16 · AI · Google APAC Cohort 1

Three of APAC's hardest AI problems.
One stack. Forty-four tests.

Emergency triage. Smart farming. Student analytics. Three different teams across the region were reinventing the same plumbing for the same kinds of problems. So I built one stack that handles all three, deployed each as its own Cloud Run service, and proved it with 44 passing tests.

Track 01 · Triage

FirstResponse AI 911 emergency triage

"Male, 40s, found under rubble, not breathing." Reply in under 2 seconds.

An ADK agent on Cloud Run that applies the START triage protocol to free-form emergency descriptions. Returns severity, immediate actions, transport priority, resource estimates. Built for non-medical volunteers in mass casualty events.

150,000+ die yearly from delayed triage 4-tier START classification (RED · YELLOW · GREEN · BLACK) 10/10 tests passing

Track 02 · Farming

AgroAdvisor Smart farming advisory

"Brown spots on my rice." Diagnosed with a confidence score and the next dry spraying window.

An ADK agent using Model Context Protocol to talk to OpenWeatherMap. Correlates real-time weather with crop symptoms to produce a diagnosis, treatment plan, and the optimal 2-3 day dry window for spraying. 20 crops, 6 growth stages, 35+ APAC locations pre-mapped.

$50B annual crop loss to preventable disease 500M smallholder farmers worldwide 16/16 tests passing

Track 03 · Education

EduPulse Student analytics in plain English

"Which students have attendance below 60% and are failing?" SQL writes itself, summary explains itself.

AlloyDB stores 200 students, 15 subjects, 2,388 enrollments. Teachers ask English questions. Gemini converts to SQL, AlloyDB executes, Gemini summarises. Zero SQL knowledge needed. Connection pooling, 12 indexes, regex-based DELETE/DROP/UPDATE blocking.

1.2M Indian students drop out yearly 5+ hours teacher time saved weekly 18/18 tests passing
3Cloud Run services in production
44Automated tests passing
1Shared ADK + Gemini stack
2.5Gemini Flash, model
0Frontend dependencies

The unifying idea

Three problems.
One shared spine.

Each track has its own UI, its own dataset, its own agent prompt. But the bones are the same: ADK orchestrates, Gemini reasons, Cloud Run serves, MCP wires up tools. Build one well, ship three.

TRACK 01 FirstResponse TRACK 02 AgroAdvisor TRACK 03 EduPulse SHARED STACK ADK · Gemini 2.5 MCP · AlloyDB · Run asia-south1 Production

Reasoning

Gemini 2.5 Flash

Agent runtime

Google ADK

Tools

Model Context Protocol

Hosting

Cloud Run · asia-south1

Proof · Live deployments

Not a slide deck.
Three live URLs.

firstresponse-ai-409924770511.asia-south1.run.app

FirstResponse AI · Cloud Run

Dark ops command center, animated radar sweep, four preset emergency scenarios, structured triage in under 2 seconds.

agroadvisor-409924770511.asia-south1.run.app

AgroAdvisor · Cloud Run

Botanical field guide UI, weather-aware spraying timeline, confidence meters, leaf-shaped tags. MCP weather tool live.

edupulse-409924770511.asia-south1.run.app

EduPulse · Cloud Run + AlloyDB

Constellation dashboard, syntax-highlighted SQL, dual-AI pipeline, 200 students seeded for instant exploration.

github.com/divyamohan1993/apac-genaiacademy

Source · Open repo

Full code, three Dockerfiles, three Cloud Build configs, 44 pytest tests, three submission decks. Apache-style organisation.

Three production AI systems in one cohort. Hire the engineer who shipped them.

I design and ship multi-agent systems on Google Cloud. ADK, Gemini, MCP, AlloyDB, Cloud Run, full CI. Same stack works for your hardest problem.