From Data Analyst to AI intelligence Tool Builder: How We Built an AI Earthquake Response Platform in 2 Days

A story from Google Cloud’s Gen AI Academy APAC Edition, Cohort 2

In March 2025, a magnitude 7.7 earthquake hit Mandalay, Myanmar. Within minutes, millions of people were online searching for answers. What they found was raw seismic data — coordinates, magnitudes, timestamps. No context. No guidance. Nothing in a language most of them could act on in a moment of panic.

I remember that feeling clearly, because Myanmar is my home. And a year later, when I joined Google Cloud’s Gen AI Academy APAC Edition as part of Cohort 2, that memory is exactly what my team and I decided to build against.

Why disaster response, why earthquakes

The Academy’s hackathon theme was broad: AI for Better Living and Smarter Communities. Teams could tackle mobility, healthcare, education, sustainability, public services — almost anything that touches everyday life. My team, Team KODA, chose Disaster Response and Recovery, and narrowed in on one specific, brutal problem: what happens to a community in the first hours after a major earthquake, when raw data exists but nobody can turn it into a decision fast enough.

That’s not a hypothetical for us. It’s something our region has lived through.

The Learn → Build → Compete journey

Going into the Academy, I wasn’t a beginner to Google Cloud. I already used tools like BigQuery in my day job as a data analyst. What the Academy actually changed was how I used them — it pushed me from running queries to shipping a live, working product, under real hackathon pressure, with mentors and a community of other builders solving their own local problems around us.

That shift — from analyzing data to being responsible for what a user does with it — is a different kind of pressure. It’s one thing to write a query that answers a question correctly. It’s another to build something that a stressed, scared person might actually rely on during a disaster. That responsibility shaped almost every decision we made.

Introducing QuakeSense

QuakeSense is what we built: an AI-powered earthquake decision-intelligence platform that turns raw seismic data into guidance people can actually use — before, during, and after an event.

A few things it does:

Live Monitor — every earthquake from the last 7 days, worldwide, pulled directly from USGS, plotted on a map where magnitude sizes the marker and tectonic plate boundaries show why a region is active. For any significant event, one click generates an AI-written community briefing — plain language, matched to real severity, not hype.

My Area — pick any country and town, and get a risk profile built from that area’s actual 50-year seismic record. It works in 8 languages, including Burmese and Hindi, because the people who need this information most often aren’t being served by English-only tools.

Ask the Data — the part I’m proudest of. It’s a natural-language agent that answers any earthquake question, but it never guesses. Ask it something like “how many magnitude 6+ earthquakes hit Myanmar since 1990?” and it converts that into real SQL, runs it against roughly 86,000 verified USGS records in BigQuery, and shows you both the query and the actual rows it pulled. We tested a generic chatbot on the same question once — it invented an earthquake that never happened. QuakeSense can’t do that. Every number is checkable against usgs.gov.

Anomaly Watch — compares this week’s seismic activity in every region against that region’s own historical average and flags what’s statistically unusual, with an AI explanation of the pattern and calm, practical context — never a prediction, because earthquakes can’t be predicted. We say that explicitly, on every page.

Response Toolkit — built for the people actually responding to an event: one click generates a formal situation report in the format emergency operations centers use, safety do’s-and-don’ts in 8 languages, and the nearest hospitals, fire stations, and verified hotlines, located from the event’s real coordinates rather than a static list.

Under the hood, it’s Google Cloud end to end: BigQuery as the analytical backbone over real, verified seismic history, Vertex AI (Gemini) for the natural-language and briefing layers, and Cloud Run serving the whole thing as a live, public Streamlit application — not a slide deck, not a local demo.

The part I care about most: trust

It would have been easy to make QuakeSense feel more “magical” by letting the AI answer freely. We deliberately didn’t. Every AI-generated number in the app is backed by a real, shown SQL query against real data. Every page that touches forecasting explicitly discloses that it does not predict future earthquakes. In a disaster tool, a confident wrong answer is worse than no answer at all — so we designed trust into the architecture, instead of just asking users to assume it.

Try it

QuakeSense is live and public:

• Live app: quakesense-537926118329.asia-southeast1.run.app
• Code: github.com/EthannMK/quakesense

Where I am now

Two days. That’s roughly how long it took to go from an idea shaped by a memory of the Mandalay earthquake, to a live product that anyone, anywhere, can open in a browser and use in their own language. Building it didn’t just teach me new Google Cloud features — it changed how I think about what “using AI effectively” actually means: not a smarter query, but a tool someone can trust when it matters.

Thanks to Hack2Skills and Google Cloud for building the Gen AI Academy APAC ecosystem, and to my teammates on Team KODA for building QuakeSense with me.

QuakeSense was built for Google Cloud’s Gen AI Academy APAC Edition, Cohort 2, under the “AI for Better Living and Smarter Communities” theme, Disaster Response & Recovery track.