WateLetric

Ignorance due to Absence leads to Expense. Save what flows.

Comment

Inspiration

The inspiration comes by looking at the mid sized companies who struggle to supervise the wastage of electricity and water usage due to uncontrollable reasons. So, the owner/boss becomes ignorant due to his times of absence in non-working hours and the electricity and water usage goes out of hand at times. This causes unnecessary expenses to the company.

What it does

WateLetric is integrated a machine learning model we built using XGBoost and a mock data of 30days. We used this model to predict the electricity and water usage in hours to come. In the graphs that is displayed, you can see the current electricity/water usage and historical usage and the usage prediction. We also simulated real time data processing, like when IoT Sensor is sending data to the system. Then, the model and Gemini will reason out and sends alert message in the dashboard and to the user phone via Telegram whenever the electricity and water usage is spiked(anomaly). There is AI chatbot powered by Gemini that allows the user to ask questions about the electricity and water usage. We also used current electricity and water usage rate in Malaysia to do the estimation of bill per month.

How we built it

Our coding agent is Google Antigravity. UI designed using Google Stitch.

  1. Backend Server:Python 3 (using standard http.server.ThreadingHTTPServer), sqlite3 relational database, urllib for API requests.
  2. Frontend UI:HTML5, CSS3, Tailwind CSS, Google Hanken Grotesk & JetBrains Mono Fonts, Material Symbols, and dynamic SVG/Chart.js rendering.
  3. Machine Learning Pipeline: XGBoost Regressor (xgboost), pandas for data manipulation, numpy for scientific calculations, and scikit-learn for evaluation metrics.
  4. Cloud Integrations:
  5. Google Gemini API: gemini-2.5-flash` model for context-aware AI assistant responses.
  6. Firebase Authentication: Firebase client SDKs v10 for login and session state routing.
  7. Telegram Bot API: HTTP webhooks for real-time Markdown alert notification dispatching.

Challenges we ran into

  1. Integrating Gemini API Key
  2. Some mistakes in the system design, but fixed it later.
  3. Difficult to find real data.
  4. Sleep and cold environment.

Accomplishments that we're proud of

  1. We successfully built a working system. Once a doubt, now we did it. Regardless of the results, we did it.
  2. We learned a lot and what we learned in class(some stuffs) actually made sense.

What we learned

  1. Using AI Tools
  2. Machine Learning
  3. If no real data, simulate and make fake test case.
  4. Think like you are in need. ## What's next for WateLetric
  5. More data
  6. Make it more scalable.
  7. Need refine the model.

Just in case, if youtube link can't open, i have attached the google drive link for video in try it out link section by adding another link section.

Built With

Share this project:

Updates