Inspiration

I🔹 Inspiration

In my hostel room, I always try to be conscious about energy usage — switching off fans, lights, and avoiding waste. But my roommate often leaves the fan running at level 5 all day, keeps the door closed while drying clothes inside, or forgets to switch off appliances when leaving.

At the end of the month, we both split the electricity bill equally, even though our consumption habits were very different. This felt unfair and highlighted a real problem:

People don’t see their real-time energy usage.

The impact comes only at the end of the month when the bill arrives.

This motivated us to design an app that makes energy visible like a phone battery %, creating awareness and nudging users to reduce waste.

What it does

AI for Energy Forecasting: How Time Series Foundation Models (TSFM) can generalize across different buildings to forecast usage without retraining.

Behavioral Science + Tech: Technology alone doesn’t save energy; visualizations and psychology can change human behavior.

Rapid Prototyping: Building a minimum viable app (MVP) with 2–3 clean screens is more effective to cover everything.

Deployment: Learned to use Streamlit, FastAPI, Firebase and integrate them for a quick end-to-end demo.

How we built it

​Frontend: Streamlit for prototype UI (dashboard, alerts, recharge screen).

Backend: FastAPI (Python) serving the AI prediction endpoints.

Database: Firebase for authentication (meter ID login) and user data storage.

AI Models: Pre-trained TSFM (Time Series Foundation Models) for load forecasting + anomaly detection.

Deployment: Hugging Face Spaces for demo hosting, GitHub for version control.

Mathematically, we modeled the energy percentage left as:

𝑃

𝑡

𝑓 𝑟 𝑎 𝑐 𝑈 𝑏 𝑎 𝑠 𝑒 − 𝑈 𝑡 𝑈 𝑏 𝑎 𝑠 𝑒 𝑡 𝑖 𝑚 𝑒 𝑠 100 P t ​

= fracU base ​

−U t ​

U base ​

times100

Where:

$U_{base}$ = baseline monthly units (e.g., 200 units)

$U_t$ = cumulative units consumed up to day $t$

$P_t$ = remaining percentage shown to the user

If $P_t < 20\%$ before day 20, the system triggers an early warning alert.

Challenges we ran into

Data Limitations: Real building consumption datasets were hard to access, so we simulated and used open datasets.

Model Adaptation: Fine-tuning TSFM for different building types required experimentation.

Design Simplicity: Converting complex analytics into a simple battery-style gauge without losing meaning.

Time Constraints: Balancing AI model integration and building a usable UI within hackathon deadlines.

🔗 Try it out " sorry we didn't record any vedio to explain - We are pretty confident on very own Idea to execute in 100% in further process"

Accomplishments that we're proud of

Turning a daily life frustration into a meaningful project idea.

Designing a unique psychological model for energy saving.

Creating a scalable architecture (from a single room to malls/factories).

Preparing a clear plan for app development with AI + real-time monitoring.

What we learned

How AI forecasting models (TSFM) can generalize to unseen buildings and predict usage without retraining.

The importance of user psychology in energy-saving — technology alone is not enough; visualization changes behavior.

Building a minimum viable product (MVP) with 2–3 clean screens is better than a complex unfinished prototype.

Deployment matters: we learned how to use Streamlit/FastAPI + Firebase for quick prototyping.

What's next for EcoWatt: Smarter Energy Monitoring

Build a working prototype using Streamlit or React Native for UI.

Connect with IoT smart meters for real-time data.

Train/fine-tune TSFM models on building datasets for accurate forecasting.

Deploy backend on AWS/Firebase with scalable APIs, or IBM cloud if it comes at affordable .

Built With

• anomaly
• cloud
• colab
• detection
• experimentation
• face
• fine-tuning
• forecasting
• foundationdb
• github
• google
• hugging
• logs)
• meter-id
• ml
• pre-trained
• python
• pytorch
• scikit-learn
• series
• spaces
• tasks
• time
• training
• tsfm
• usage