Ever had a device just... stop working, and you had no clue which part actually gave up? That's the itch I built DiagnostiX AI to scratch.
It's a small machine-learning web app: you tell it the device, how old it is, how you use it, and the symptoms you're seeing — and it points you to the hardware component most likely at fault, so you know where to start looking.
👉 Click here to open DiagnostiX AI
Just click the link, pick your device, enter the symptoms — and the model runs right in your browser.
⚠️ It's a guide, not a guarantee. DiagnostiX shows you the most likely culprit. Always confirm with a real hardware check before replacing anything.
Open the link above and the app loads straight in your browser — no setup, no install, nothing to download. Choose your device, enter three symptoms, and get an instant diagnosis.
- Works across phones, laptops, desktops, tablets, smartwatches, routers, smart TVs and printers
- Predicts the most likely faulty component, plus a couple of backup suspects
- Tells you how strongly your symptoms match the prediction
- Only offers symptom combinations the model actually understands — so you never get nonsense inputs
- You enter the device, its age, daily usage, when the trouble started, and three symptoms.
- Behind the scenes, the text fields get one-hot encoded and the numbers pass straight through (
ColumnTransformer). - A Random Forest (300 trees, class-balanced) makes the call, all wrapped in one tidy scikit-learn
Pipeline. - You get the top suspect, two alternatives, and a match-strength bar.
The model, the preprocessing, and the list of valid inputs all travel together in a single artifact — so the app and the model never drift out of sync.
The dataset behind DiagnostiX is something I gathered from the ground up, rather than downloading off the shelf:
- A Google Form survey where everyday users described the problems they'd run into with their devices and what the fault eventually turned out to be.
- Inputs from nearby hardware shops, service centres and repair technicians, who shared the symptom-to-fault patterns they deal with every day.
- Online problem reports — common device issues that people post and discuss across the web, distilled into the same symptom-to-fault structure.
Combining real user reports, online complaints, and hands-on repair experience is what makes the predictions feel grounded in how devices actually fail, instead of being made-up numbers.
A small honesty note: because the collected data is fairly clean and the symptom patterns map quite consistently onto faults, the model scores very high on paper. Real-world symptoms are messier and more ambiguous, so treat the prediction as a strong starting hint — not a final verdict.
To keep the evaluation honest, train.py runs:
- a stratified train/test split
- 5-fold cross-validation
- per-class precision / recall / F1
- a confusion matrix you can actually look at
pip install -r requirements.txt
python train.py # trains + prints the full metrics (optional)
streamlit run app.py # the app trains itself on first run if no model exists- Make sure the dataset CSV is in the repo.
- Go to https://share.streamlit.io → New app.
- Pick this repo, branch
main, main fileapp.py. - (Optional) Choose Python 3.12 under Advanced settings.
- Hit deploy — the app trains the model on first load and caches it.
- Keep growing the dataset with more survey responses and repair-shop records
- Understand free-text symptoms using fuzzy matching / embeddings
- Add confidence calibration and an honest "I'm not sure — get it inspected" answer
- A feedback loop so every confirmed repair makes the model a little smarter
- Surface repair cost & time estimates alongside the predicted fault
DiagnostiX-AI/
├── app.py # the Streamlit interface
├── diagnosis.py # loads/serves the model + prediction logic
├── train.py # training, evaluation, metrics
├── tests/ # unit tests
├── requirements.txt
└── *.csv # the training data
Ashwin Dubey — ECE student at Chandigarh University, learning by building.
- GitHub: https://github.com/That-ASHWIN
- LinkedIn: https://www.linkedin.com/in/ashwin-dubey-b27657302/
- Email: cuniversity223@gmail.com