DUVVURI DURGA PRASAD

I began by designing the layout with semantic HTML5 and styled it using modern CSS3, focusing on a clean and responsive layout. JavaScript was used to handle the dark/light theme switch, smooth navigation, and dynamic content rendering for projects.

The website was version-controlled with Git and deployed using GitHub Pages for free and fast hosting. All project sections link directly to live demos and source code for transparency and easy access.

Key highlights:

• Dark/light theme toggle
• Fully responsive design
• Integrated project showcase with live links
• Deployed using GitHub Pages

I started by collecting and cleaning movie metadata to create meaningful features for similarity comparison. Using TF-IDF vectorization, I converted textual features into numerical vectors. Cosine similarity was then used to compute similarity scores between movies.

The TMDB API was integrated to dynamically fetch high-quality movie posters to display with recommendations, improving visual appeal and user engagement.

The entire system was wrapped into an interactive web app using Streamlit, providing a simple and responsive UI that allows users to select movies and instantly view recommendations with posters.

Key highlights:

• Advanced feature engineering combining multiple metadata features
• Real-time API integration for poster retrieval
• End-to-end deployment with Streamlit for an interactive user experience

I structured the project as an end-to-end ML pipeline, starting with data ingestion and preprocessing, followed by feature engineering, model training, and evaluation.

Using DVC, I version-controlled the datasets and pipeline stages, enabling reproducibility and easy collaboration. MLflow tracked experiments including model parameters and evaluation metrics, facilitating model comparison and selection.

The best model was wrapped in a Flask API for serving predictions, while the entire setup was containerized using Docker to ensure environment consistency. Deployment was automated on AWS EC2, and GitHub Actions CI/CD pipelines ensured code quality and smooth updates.

Key highlights:

• Modular pipeline components enabling flexible experimentation.
• Seamless experiment tracking and model registry with MLflow.
• Robust deployment strategy with Docker and AWS.
• CI/CD automation via GitHub Actions for testing and deployment.

I began by cleaning and preprocessing a large dataset of restaurant orders. Using exploratory data analysis and visualization techniques (Seaborn, Plotly), I identified key trends such as the popularity of American and Japanese cuisines, and how delivery efficiency correlates with customer satisfaction.

Built predictive models for delivery time, order cost, and customer ratings using relevant features like cuisine type, preparation time, and day of the week.

The insights help stakeholders improve operational efficiency, predict delivery delays, and enhance customer experience through data-driven decisions.

Key highlights:

• Comprehensive exploratory data analysis revealing customer behavior patterns.
• Predictive modeling for critical operational metrics.
• Interactive visualizations supporting decision-making.

Data preprocessing and feature selection were performed to identify the most relevant predictors of exam outcomes. I trained multiple machine learning algorithms, tuning hyperparameters to maximize predictive accuracy.

An interactive web application was created using Streamlit to allow users to input student data and instantly receive performance predictions.

Key highlights:

• Feature engineering for improved model accuracy.
• Model selection and tuning to identify optimal predictors.
• Easy-to-use Streamlit app for real-time predictions.

I trained a Random Forest classifier using cleaned loan application data and stored the model using pickle. The Flask app collects user inputs via Bootstrap-powered responsive forms and returns approval predictions.

Predictions and user contact messages are saved to an SQLite database for persistence. Front-end design was enhanced with custom CSS and JavaScript for usability.

Project Structure Highlights:

• app.py: Flask backend application
• templates/: HTML pages including forms and result views
• static/: CSS, images, and JavaScript files
• random_forest_model.pkl: Serialized model file
• schema.sql: SQLite database schema

I performed extensive exploratory data analysis using boxplots, KDE plots, and scatter matrices to understand feature distributions and relationships. Feature engineering improved model inputs, while missing data was carefully handled.

I trained and tuned multiple classifiers, ultimately selecting the Random Forest model based on performance metrics. The final model was saved for future use and deployment.

Key highlights:

• Thorough data exploration and visualization.
• Comparative evaluation of different machine learning models.
• Robust feature engineering and model selection.

I developed the machine learning pipeline by training a Random Forest classifier on a labeled breast cancer dataset, tuning it for high accuracy and reliability.

The model was serialized using pickle and integrated into a Flask web app that allows users to input tumor features via a clean and responsive HTML form.

Static assets like CSS and JavaScript were used to enhance the UI/UX. The app immediately provides a prediction on submission, allowing for quick cancer risk assessments.

Project Structure Highlights:

• app.py: Main Flask application handling routes and prediction logic
• model.py: Contains the model training and loading functions
• classifier_rf.pkl: Serialized Random Forest model
• templates/index.html: User input form and result display
• static/: CSS and JavaScript files for front-end styling and interaction

Prerequisites: Python 3.x, Flask, scikit-learn, NumPy, Pickle.

Explored structured customer datasets and engineered features such as income-to-age ratio and employment status. Trained classification models with hyperparameter tuning and evaluated using ROC AUC and confusion matrix analysis.

Implementation Process:

• Processed structured demographic and financial data
• Engineered key features like income-to-age ratio
• Trained Decision Tree and Logistic Regression models
• Evaluated with ROC AUC and confusion matrix