Predictify - Urban Sound Classifier

Built an end-to-end ML product where users upload audio and receive a predicted urban sound class with a confidence score.

Problem

I wanted to move beyond tutorial ML examples and build something that combined real model work, a clean API contract, and a usable web interface.

Solution

Trained the classifier on UrbanSound8K features, exposed predictions through Flask, then connected it to a React frontend through a Node.js proxy.

Role

Role: Full-Stack Developer and ML Engineer

Duration: 5 weeks

Responsibilities

• Model training and evaluation
• Prediction API with Flask
• Node proxy for uploads and request handling
• React frontend with result states

Highlights

• File validation across every layer
• Cross-service coordination between Python, Node, and React
• Fast prediction feedback for uploaded audio

Outcome

• Reached about 85 percent accuracy on the test split
• Typical end-to-end prediction time stayed under two seconds
• Showcased ML engineering alongside full-stack delivery

What I Learned

• Feature engineering matters as much as model choice
• Cross-language systems need clear contracts before implementation

Stack

React · TypeScript · Node.js · Express · Python · Flask · Scikit-learn · Librosa