Inspiration
Energy consumption data is everywhere — yet households and businesses often lack visibility into how their power is being used or wasted. With rising energy costs and growing sustainability goals, we wanted to build something that transforms raw meter data into actionable intelligence. Powerlytics was inspired by the idea of turning everyday power usage into insights that empower greener decisions — all through the power of AI and cloud data.
What it does
Powerlytics is an AI-powered energy optimization platform that collects real-time smart meter data, automates ingestion via Fivetran, and stores it in BigQuery for large-scale analytics. From there, Vertex AI models forecast demand spikes, detect anomalies, and suggest efficiency improvements.
Users access an interactive Next.js dashboard that visualizes usage patterns and includes a conversational AI assistant (powered by Gemini) that can answer questions like:
“Why did my usage spike yesterday?” “How can I reduce peak-time consumption?”
How we built it
Data Ingestion: Smart meter data streams through a custom Fivetran connector built with the Connector SDK.
Data Warehouse: Data lands in Google BigQuery, where it’s transformed into analytics-ready tables.
Machine Learning: Vertex AI trains models for consumption prediction and anomaly detection.
Backend: A FastAPI service handles API requests, runs Vertex predictions, and queries BigQuery.
Frontend: Built with Next.js, featuring charts (Recharts.js) and a chat UI integrated with Gemini APIs for natural-language analysis.
Deployment: Containerized with Docker, hosted on Cloud Run, fully integrated with Google Cloud IAM and logging.
Challenges we ran into
Setting up a reliable streaming ingestion pipeline with the Fivetran SDK and aligning schemas for BigQuery.
Handling latency and cost optimization when querying large energy datasets.
Configuring Vertex AI endpoints securely and managing authentication tokens for real-time inference.
Designing a dashboard that remained intuitive while visualizing complex energy trends.
Integrating a chatbot-style interface that could interpret natural language queries into meaningful BigQuery requests.
Accomplishments that we're proud of
Built a fully functional end-to-end pipeline from IoT data to user dashboard in under 48 hours.
Created a custom Fivetran connector that streams mock smart meter data automatically.
Deployed a Vertex AI forecasting model that achieved >92% accuracy on short-term usage prediction.
Designed a sleek Next.js dashboard that merges analytics and conversational AI into a single experience.
Showcased the power of Google Cloud integration — from ingestion to ML to visualization — in one cohesive system.
What we learned
How to orchestrate multi-service pipelines across Google Cloud, Fivetran, and containerized apps.
How Vertex AI simplifies model training and deployment compared to traditional ML stacks.
How important schema design and ETL workflows are for large-scale analytics.
That making AI interpretable to users (via chat UI + visualization) dramatically improves engagement.
What's next for Powerlytics
Scale to real IoT data by integrating with smart meter APIs and renewable energy sources.
Add Elastic hybrid search for querying across historical and real-time energy data.
Develop AI-powered alerts for unusual consumption and automatic efficiency recommendations.
Launch a mobile app version of the dashboard for real-time monitoring.
Partner with utilities and sustainability organizations to turn Powerlytics into a deployable green tech solution.
Log in or sign up for Devpost to join the conversation.