clear home.
Built with Google Cloud

The cloud behind Clear Home.

Housing search, helpful AI, and maps.
The Google Cloud tools that bring it all together.

  • Hostingus-east4Based in Northern Virginia.
  • AI computeNVIDIA L4Private GPU for Gemma 4.
  • LocationPlaces + RoutesNearby services and travel times.
  • AccessPrivate by designThe GPU accepts requests from our website.

How It Fits Together

From your first question to a helpful answer.
One connected system, built on Google Cloud.

01 Your device

Your browserExplore housing and maps with Maps JavaScript.

02 Clear Home

Cloud RunOur Flask website and API.
clear-home · us-east4

03 Google services

Cloud Run GPUPrivate Gemma 4 26B on NVIDIA L4.
Gemini APIAI fallback and Google Search grounding.
Google Maps PlatformNearby places and walking routes.

Supporting every release

Secret ManagerStores Gemini and Maps keys.
Cloud BuildBuilds releases from our Dockerfile.
Artifact RegistryStores website and GPU images.

What Each Service Does

And what we do to push it further than the defaults.

Compute

Cloud Run

Hosts the whole website and its API as one container.

  • Close to youRuns in us-east4, the Google region closest to Northern Virginia.
  • Costs stay lowShuts down when nobody is visiting and never runs more than two copies.
  • Photos load fastListing photos are stored inside the app, so pages don’t wait on a landlord’s website. A photo we couldn’t download is left off and the listing still shows.
  • Instant answersCommon questions are answered when the server starts, so the resources page responds right away.
  • Data stays safeThe live site is read-only. Data changes are reviewed on our computers first and shipped as code.
GPU

Cloud Run GPUs

Our own NVIDIA L4 graphics card runs Gemma 4 26B (through Ollama), just for Clear Home.

  • Quick repliesEveryday questions skip hidden “thinking”, so answers start sooner (we haven’t published timed measurements yet). Questions with specific needs (pets, furnished, roommates…) get time to check every home first.
  • PrivateIt refuses public calls. The website signs each request with its own Google identity.
  • Pay only when usedIt shuts down when idle. By Google’s list prices for an L4 on Cloud Run (checked October 2026) it costs roughly $1 an hour while awake; that’s an estimate, not a measured bill. During judging, one copy stays on so the first answer doesn’t wait for a cold start.
  • Ready before you askThe GPU starts waking as soon as the chat scrolls into view. If it isn’t ready when you send, the answer comes from the Gemini API instead.
  • Falls back, step by stepIf the GPU is asleep or slow, the chat switches to the Gemini API after 20 seconds. If that fails too, you get an answer built from the reviewed list without AI, plus the 211 helpline.
AI

Gemini API

Runs Gemma 4 (gemma-4-26b-a4b-it) and Gemini with Google Search.

  • Understands youGemma turns your message into search filters, then writes the answer using only verified listings, shown as it types.
  • Finds new placesGemini with Google Search looks for listings we don’t have yet. Only places that appear in the search results are kept, and the map marks them “not verified”.
  • Removes stale listingsGemma rereads listing pages and hides ones that are no longer offered. It can hide listings, never add them.
  • Protects your infoSocial Security and account numbers are removed before any text is sent.
Maps

Google Maps Platform

Three Google Maps tools: Maps JavaScript, Places API and Routes API.

  • The mapMaps JavaScript draws the map with a pin for every verified home.
  • Places nearbyPlaces API finds the place you name (an employer, a hospital, a school), plus the supermarkets, pharmacies and transit stops near a home.
  • Real travel timesRoutes API gives real drive and transit times from the closest homes, and true walking times to the store.
  • Keys kept apartThe server and the browser use separate keys, so the server key never reaches a web page.
Security

Secret Manager

Holds the Gemini key and both Maps keys.

  • Never in the codeKeys are not in the code, the repository or the app image.
  • Easy to changeSwapping a key means adding a new version and redeploying, with no code change.
Delivery

Cloud Build and Artifact Registry

Turn our code into running containers.

  • Fixes go live fastEvery deploy builds a fresh container from our Dockerfile, so a pushed fix usually goes live 4–6 minutes after the build starts (Cloud Build’s typical time for this app).
  • Room for the big AI imageThe 17 GB Gemma GPU image is built on a larger Cloud Build machine and stored in Artifact Registry next to the site.

Full Power, Kept in Check

Using a GPU and live AI on a hackathon budget means building in limits from day one.

Pay only while in use

The website scales to zero and is capped at two copies. The GPU runs at most one copy and can scale to zero.

One switch to stop GPU costs

Removing one setting sends every chat back to the Gemini API, with no code change.

Nothing public that shouldn’t be

The GPU is private, keys live in Secret Manager, and the hosted site can’t change its own data.

See It Working

Ask Gemma about housing, rent help or shelters in Northern Virginia and Blacksburg.

Get Started for Free