🏙️ First Runner — Conquer the City by Running
The Problem: The Sedentary Epidemic in the Digital Age Modern fitness apps are static text boxes and boring graphs. They track your run, but they don't engage you. Motivation fades. We created First Runner to gamify the physical world, turning your daily route into a strategic turf war driven by a hyper-responsive, real-time AI coach powered by Google Gemini.
The Solution: A Cyberpunk Turf War First Runner is a gamified, location-based fitness app where users conquer real-world city territories by physically running, walking, or cycling.
• Territory Dominion: Every mapped street is a grid ready to be claimed. Paint the city with your clan's colors. • The Oráculo (Live AI Coach): Our crown jewel — powered by Gemini 2.5 Flash Native Audio. It talks to you during the run, handles barge-in interruptions mid-sentence, knows your real-time GPS pace and territory status, and responds in under 1 second.
─── How It Maps to the 3 Challenge Tracks ───
🎙️ Live Agent (Primary Track) The Oráculo is a continuous voice agent that listens, responds, and handles real interruptions during a live run. No turn-based delays. Context-aware via real-time GPS telemetry injection into the Gemini session.
🎨 Creative Storyteller After each run, the Oráculo delivers a "Mission Debrief" — a multimodal narrative mixing epic text, map snapshots, and performance stats in a single interleaved output stream.
🖥️ UI Navigator The agent receives screenshots of the tactical map HUD, interprets red zones (enemy territories), and alerts the runner about visual changes in real-time.
─── The Architecture ─── The Flutter app captures raw 16kHz PCM audio and streams it via WebSocket to our Node.js bridge on Google Cloud Run. This bridge orchestrates a bidirectional Gemini 2.5 Flash Native Audio session. Simultaneously, the app silently injects live GPS telemetry (pace, distance, territory status from Firestore) into the context — making every response tactically relevant.
─── Tech Stack ─── • Google GenAI SDK + Gemini 2.5 Flash Native Audio — live voice coach • Google Cloud Run — WebSocket backend • Firebase Firestore + Auth — real-time territory sync • Flutter (Dart) — cross-platform mobile frontend • OpenStreetMap + Nominatim — map tiles and geocoding (third-party) • WalletConnect + web3dart — blockchain achievements (third-party)
─── Key Challenges Solved ─── • Barge-in without feedback loop: AudioContextAndroid + gainTransientMayDuck + PCM buffer management • Zero-latency telemetry injection: silent JSON context updates without breaking the audio stream • Resilient WebSocket reconnection during runs in low-signal areas
─── Findings & Learnings ─── Native Audio was the game-changer. Bypassing TTS made the Oráculo feel like a living entity, not a script. We learned that grounding the AI with real GPS context — territory status, pace, rival positions — transforms generic motivation into genuine tactical coaching. The future of interaction is continuous, multimodal presence.
─── Business Model ─── B2B physical traffic platform: gyms, pharmacies, and retailers become official pit stops. The Oráculo guides runners to partner locations via contextual audio missions.
Built for the Gemini Live Agent Challenge 2026 | #GeminiLiveAgentChallenge
📝 Behind the build: https://medium.com/@tpd.egon89/como-construí-um-jogo-de-corrida-cyberpunk-com-gemini-live-ai-e-o-que-aprendi-no-caminho-9c53b5db9577
Built With
- dart
- firebase
- firestore
- flutter
- gemini-api
- gemini-live
- google-cloud-run
- javascript
- node.js
- openstreetmap
- web3
- websockets


Log in or sign up for Devpost to join the conversation.