UrjaBandhu

UrjaBandhu: Professional power, optimized every hour

Comment

⚡ UrjaBandhu – Smart Home Electricity Optimizer

🌱 Inspiration

In India, millions of households face a dual challenge: rising electricity costs that strain family budgets and a growing national carbon footprint. The core of this problem is an information gap—most people don't know which specific appliances are driving their high bills. While smart solutions exist, they are often too expensive, complex, or fail to provide the granular, actionable insights needed to make a real difference.

This inspired us to create UrjaBandhu. We envisioned a system that doesn't just monitor energy but empowers users to actively control it. Our platform is engineered on a unique trifecta of innovation:

  • 💸 Budget-Aware AI: Proactively manages consumption to meet financial goals.
  • 🔍 Granular Disaggregation: Pinpoints the energy usage of individual appliances without expensive hardware.
  • 🗣️ Inclusive Voice Guidance: Makes energy insights accessible to everyone, regardless of technical literacy or language.

By integrating these features, UrjaBandhu transforms energy management from a guessing game into a clear, data-driven process, making it accessible, actionable, and affordable for every household.

💡 What It Does

UrjaBandhu is a smart, AI-driven electricity management system. In essence, it acts as a micro-SCADA system for the home, providing centralized intelligence and control over household energy consumption. It:

  • 📷 Detects appliance-level usage via OCR + NILM.
  • 📊 Forecasts monthly consumption and bills with time-series models.
  • 🔔 Sends budget alerts and personalized energy-saving tips.
  • 🗣️ Supports multi-language TTS/STT notifications.
  • 📌 Generates actionable insights, like: > "Your fan running 8 hours a day will add an extra ₹120 to this month's bill."

🛠️ How We Built It

💻 Tech Stack

  • Frontend: Next.js + Tailwind CSS (mobile-friendly dashboard)
  • Backend: FastAPI + Supabase (PostgreSQL + real-time subscriptions)
  • Deployment: Docker containers
  • Analytics: InfluxDB for time-series data
  • Voice Services: Azure Cognitive Services for TTS/STT

🧠 Machine Learning Models

🔌 NILM Disaggregation (LSTM/HMM)

We estimate appliance-level usage from the total household consumption. The total power draw is the sum of the power draw of individual appliances.

$$ U_t = \sum_{i=1}^{N} u_{i,t} $$

Where:

  • \(U_t\): Total household consumption at time \(t\).

  • \(u_{i,t}\): Usage of appliance \(i\) at time \(t\).

📈 Time-Series Forecasting

We predict future consumption by learning a weighted sum of past consumption values.

$$ X_t = c + \sum_{i=1}^{p} \phi_i X_{t-i} + \varepsilon_t $$

Where:

  • \(X_t\): The predicted consumption at the current time.

  • \(X_{t-i}\): Past consumption values.

  • \(\phi_i\): Weights learned by the model.

  • \(\varepsilon_t\): The random error term.

💰 Budget Optimization Constraint

This ensures that the total projected electricity cost for the month stays within the user's defined budget.

$$ \sum_{t=1}^{T} P_t \cdot U_t \leq B $$

Where:

  • \(P_t\): Price per unit of electricity at time \(t\).

  • \(U_t\): Total usage at time \(t\).

  • \(B\): The user-defined monthly budget.

🔍 OCR + NLP

  • OCR: Tesseract for reading meters and device labels.
  • NLP: GPT for device detection and generating personalized tips.

🚧 Challenges We Ran Into

  • 📉 Dealing with noisy and inconsistent data from smart meters and manual readings.
  • 🧮 Performing real-time predictions on resource-limited hardware.
  • 🧠 Ensuring ML model outputs are interpretable and actionable for everyday users.
  • 🌐 Designing a user-friendly, multilingual interface for a diverse audience.

🏆 Accomplishments We're Proud Of

  • ✅ Engineered a budget-aware optimization engine that delivers appliance-level savings tips through an accessible, multi-language voice interface.
  • 🔍 Achieved accurate appliance-level disaggregation with NILM.
  • 💬 Implemented proactive budget-aware notifications to prevent overspending.
  • 🌍 Created a multi-language voice-enabled interface for greater inclusivity.
  • 🔗 Successfully combined AI, data analytics, and UX into a scalable, real-world solution.
  • 📈 Integrated forecasting and optimization into a single, actionable platform.

📚 What We Learned

  • How to apply NILM and time-series models to messy, real-world electricity data.
  • The importance of designing human-centered interfaces for diverse user groups.
  • The trade-offs between accuracy, interpretability, and efficiency in ML systems.
  • End-to-end integration of AI, data pipelines, and full-stack engineering under tight deadlines.
  • How to translate technical innovation into tangible financial and environmental impact.
  • Methods for handling uncertainty in predictions and real-world optimization constraints.

🚀 What's Next for UrjaBandhu

  • 🔄 Automated Budget-Driven Control: Evolve from alerts to action by automatically optimizing appliance schedules to guarantee the user's monthly budget is never exceeded.
  • 📱 Mobile App: Develop a native mobile app for push notifications and richer analytics.
  • 🏘️ Community Insights: Provide community-level energy insights for shared optimization goals.
  • 🔍 Advanced ML: Incorporate models for anomaly detection, predictive maintenance, and renewable energy integration.
  • 🇮🇳 Nationwide Deployment: Scale the solution across India to maximize cost savings and carbon footprint reduction.

Built With

Share this project:

Updates