Built for the AI Partner Catalyst: Google Cloud Partnerships Hackathon | Confluent Challenge Entry

"Turning Sustainable Intent into Local Action through Data in Motion."

EcoStream is a voice-driven AI navigator designed to solve UN Sustainable Development Goal (SDG) 12: Responsible Consumption and Production.

In mega-cities like Lagos, citizens face a "Green Visibility Gap"—they want to shop sustainably but lack real-time data on local eco-friendly markets. EcoStream bridges this gap by using Vertex AI and ElevenLabs to provide a conversational guide that maps local markets and provides instant carbon-savings transparency.

Every interaction is treated as Data in Motion, streamed through Confluent Cloud and monitored for reliability by Datadog.

EcoStream is built on a high-availability, multi-partner cloud architecture:

1. Secure Entry: Google OAuth 2.0 ensures identity-based tracking.
2. The Brain: Vertex AI (Gemini 2.5 Flash) processes voice intent and calculates carbon impact.
3. The Voice: ElevenLabs (Rachel) provides natural, human-like coaching and feedback.
4. The Heartbeat: Confluent Cloud handles real-time streaming of shopping events via the market-activity Kafka topic.
5. The Eyes: Google Maps JavaScript API uses Vector Maps and Advanced Markers to navigate the user.
6. The Warehouse: Google BigQuery persists data for long-term urban sustainability analytics.
7. The Watchtower: Datadog provides full-stack observability, tracking AI latency and business impact metrics.

• Data in Motion: We moved beyond "Data at Rest." Every search produces a rich JSON payload to the market-activity topic.
• Real-time Logic: This architecture enables city planners to see sustainability pulses as they happen, not weeks later in a report.

• Application Health: Deep integration with Google Cloud Platform via the Marketplace.
• Actionable Items: Automated monitors for BigQuery quotas and custom business events for AI searches.
• Incident Management: Every AI anomaly triggers an automated Datadog Case with full user context.

• Voice Design: Utilizing the Multilingual v2 model to provide an encouraging, human-like persona (Rachel) for the Eco-Guide.
• Accessibility: Removed the barrier of typing, allowing users to interact entirely through natural speech.

• Vertex AI (Gemini 2.5 Flash)
• Cloud Run (Containerized Hosting)
• BigQuery (Data Warehousing)
• Google Maps Platform (Advanced Markers & Vector Engine)
• Google Cloud Speech-to-Text
• Cloud Identity (OAuth 2.0)
• Secret Manager
• Cloud Build
• Artifact Registry

• app.py: The production Flask backend.
• templates/index.html: The modern, voice-enabled frontend.
• ecostream_dashboard.json: [Hard Requirement] Exported Datadog Dashboard configuration.
• datadog_monitor.json: [Hard Requirement] Exported BigQuery quota monitor logic.
• Procfile & Dockerfile: Professional deployment configurations.

Our final demo features Veo 3.1 generative video, illustrating a citizen's journey from a digital search in Lagos to the physical act of sustainable shopping in a local market. Ambient audio was generated using ElevenLabs Sound Effects to provide a hyper-realistic experience.

Jeremiah Ope
Solutions Architect & AI Developer
Abuja, Nigeria 🇳🇬