AI & technology lab
Hands-on experiments that run in this page. No AI subscription or account key is needed. These educational demos do not control your appliances.
Experiment · Text classification
A small Naive Bayes model learns word patterns from labelled examples. Try “not noisy” to see a limitation: it does not understand negation or meaning. English words only; everything runs locally.
Recognised words: noisy, leaking. A label is not proof of product quality.
4/5 match the expected teaching label. These fixed examples start outside training; adding them to training removes that separation. This tiny result is not a general accuracy estimate or confidence score.
Negation, unfamiliar vocabulary and contradictory labels can fail. Try correcting the examples, then reset to compare.
Experiment · Learn a numeric relationship
Fit a straight line to example measurements. These starting numbers are synthetic, not appliance test data. Add an outlier to see how a learned model changes.
3.85 kWh
Learned line: 0.770 × hours + 0.000
Outside the training range: this is extrapolation.
This model ignores weather, standby power and compressor cycling. Use it to learn about fitting and bias, not to predict your bill.
Prompt workshop · Template, not a generative AI model
No request is sent to an AI provider. Verify any generated answer against the original sources.
Interactive machine-learning experiment
This real k-nearest-neighbours classifier votes using the three closest labelled examples. Change the conditions, then add your own example to see how the output changes. It is a small teaching model, not a validated comfort or energy controller.
6/30 examples · Changes disappear when you leave this page.
Model prediction
Fan
Votes: Cooling 1/3 · Fan 2/3
Inputs are scaled by their displayed ranges. A tied vote uses the nearest example’s label. Votes are not calibrated confidence. Missing factors such as humidity can make the answer unsuitable.
Technology experiment · arithmetic, not AI
₹192 over 30 days
Watts ÷ 1,000 × hours × 30 × tariff. Assumes constant power while running. Cycling appliances and variable-speed devices need measured average consumption.
Explore Google’s browser image-classification guide and linked demo. Classifiers choose from learned categories; they do not understand every possible object.
Open external experiment hub ↗External service: availability, permissions and privacy practices may differ. Review them before providing images or camera access.
Explore interactive machine-learning examples maintained in the TensorFlow ecosystem.
Open external experiment hub ↗External service: availability, permissions and privacy practices may differ. Review them before providing images or camera access.