MLOps/ML Engineer

Welcome to my portfolio! I am a seasoned MLOps professional with over 5 years of experience in the development and productionization of machine learning, deep learning, and generative AI solutions. Committed to innovation and impactful contributions in the AI field.

Work Experience

Senior ML/MLOps Consultant (10/2023 - Present)

• Company: Firstsource, Hyderabad, India
  • Leading the design and deployment of cutting-edge ML systems on multi-cloud platforms.
  • Showcasing expertise in innovative solutions and building robust ML pipelines.
  • Utilizing Big Data technologies like Pyspark, BigQuery, and Snowflake for actionable insights.

Senior MLOps Engineer (12/2021 - 10/2023)

• Company: Kroll, Hyderabad, India
  • Designed and executed end-to-end MLOps pipeline, accelerating project delivery by 40%.
  • Optimized ML pipelines, reducing model training time by 50%.
  • Implemented MLOps on multi-cloud and Kubernetes platforms, achieving 70% faster time-to-market.
  • Introduced automated model monitoring and alerting system, reducing model errors by 30%.

DevOps Engineer (05/2021 - 11/2021)

• Company: OSI Digital, Hyderabad, India
  • Customized solutions, reducing production delay by 70% and ensuring 95% on-time delivery.
  • Designed and maintained cutting-edge MLOps infrastructure using Azure ML Services.
  • Strengthened real-time monitoring and alerting systems, improving overall data reliability and accuracy.
  • Implemented automated deployment pipelines, resulting in a 30% reduction in deployment errors.

System Administrator (09/2019 - 04/2021)

• Company: PCS Global, Kolkata, India
  • Managed, troubleshot, and updated hardware and software assets, resulting in a 40% reduction in system downtime.
  • Implemented preventive measures, ensuring 99% system availability.
  • Automated patching and configuration processes, slashing deployment time by 50%.
  • Introduced centralized logging and monitoring system, improving incident response time by 40%.

Skills

• Machine Learning: MLflow, Kubeflow, TensorFlow, PyTorch, PySpark, H2O, Azure ML, AutoML, AWS Sagemaker, OpenCV, Keras, Mlflow
• Data Engineering: DataOps, ETL, Dagster, Airflow, Spark, Hadoop, Databricks, Data Warehouse
• DevOps: Kubernetes, Docker, Ansible, Terraform, Jenkins, CI/CD, Azure DevOps, AWS CodePipeline, Bash Scripting, PowerShell
• Databases: SQL, MS-SQL, MySQL, PostgreSQL, Cassandra, MongoDB
• Programming Languages: Python, C++, R, Java, HTML, CSS, JavaScript, YAML, JSON
• ML Models & Algorithms: KNN, CNN, ANN, RNN, Linear Regression, Logistic Regression, Decision Trees, KMeans, DBScan, Random Forests, Naive Bayes, Gradient Boosting, SVM, NLP, Langchain, LLaMA, Mistral
• Deep Learning Frameworks: TensorFlow, Keras, PyTorch, OpenCV
• Data Analysis & Visualization: Numpy, Pandas, Scipy, Matplotlib, Seaborn, Plotly, Tableau, Power BI
• Cloud Platforms: AWS, Azure, GCP
• Version Control: Git, GitLab, SVN
• Operating Systems: Ubuntu, Windows Server 2019, RedHat Linux

Education

• M.Tech: Mechanical Engineering (2019)
  • Biju Patnaik University of Technology, Odisha, India
• B.Tech: Mechanical Engineering (2016)
  • Centurion University of Technology and Management, Odisha, India

Projects

Machine Learning Projects

1. Image Classification with Convolutional Neural Networks (CNN)

• Tools Used: TensorFlow, Keras
• Description:
  • Implemented a CNN for image classification tasks using the CIFAR-10 dataset.
  • Challenges:
    • Handling limited labeled data and preventing overfitting.
    • Fine-tuning hyperparameters for optimal performance.
  • Solutions:
    • Used data augmentation techniques to address limited data.
    • Implemented regularization techniques and hyperparameter tuning.
  • Conclusion:
    • The CNN model achieved high accuracy in image classification, demonstrating its effectiveness.

2. Natural Language Processing (NLP) - Sentiment Analysis

• Tools Used: PyTorch, NLTK
• Description:
  • Built a sentiment analysis model using deep learning for processing and analyzing text data.
  • Challenges:
    • Dealing with noisy and unstructured text data.
    • Addressing class imbalance in sentiment labels.
  • Solutions:
    • Preprocessed text data with tokenization and stemming.
    • Applied oversampling techniques to balance sentiment classes.
  • Conclusion:
    • The NLP model provided accurate sentiment analysis, enhancing text data understanding.

3. Recommender System using Collaborative Filtering

• Tools Used: Surprise, scikit-learn
• Description:
  • Developed a recommendation system based on collaborative filtering techniques.
  • Challenges:
    • Handling sparse user-item interaction matrices.
    • Optimizing recommendations for real-time usage.
  • Solutions:
    • Applied matrix factorization for efficient handling of sparse matrices.
    • Implemented caching mechanisms for real-time recommendation generation.
  • Conclusion:
    • The recommender system provided personalized suggestions, improving user experience.

4. Generative Adversarial Network (GAN) for Image Synthesis

• Tools Used: TensorFlow, Keras
• Description:
  • Created a GAN model for generating realistic images, focusing on faces.
  • Challenges:
    • Training stable and high-quality GANs.
    • Managing mode collapse and diverse image generation.
  • Solutions:
    • Employed techniques like spectral normalization for stable training.
    • Fine-tuned hyperparameters to encourage diversity in generated images.
  • Conclusion:
    • The GAN model successfully generated realistic and diverse facial images.

5. Time Series Forecasting with Long Short-Term Memory (LSTM)

• Tools Used: PyTorch, TensorFlow
• Description:
  • Implemented an LSTM-based model for predicting time series data.
  • Challenges:
    • Capturing long-term dependencies in time series data.
    • Handling varying trends and seasonality.
  • Solutions:
    • Utilized LSTM layers to capture long-term dependencies.
    • Applied feature engineering techniques to handle trends and seasonality.
  • Conclusion:
    • The LSTM model provided accurate time series forecasting, valuable for various applications.

6. Consumer Finances in USA

• Description:
  • Implemented a machine learning model to analyze and predict consumer spending patterns in the USA.
• Challenges:
  • Handling diverse financial data sources and formats.
  • Addressing imbalances in labeled datasets for training.
• Solutions:
  • Developed data preprocessing pipelines for diverse data sources.
  • Applied oversampling techniques to balance the dataset.
• Conclusion:
  • The model accurately predicted consumer spending patterns, aiding financial planning and decision-making.

7. DS Admissions in WQU

• Description:
  • Created a predictive model for assessing admission probabilities based on various factors in a Data Science program.
• Challenges:
  • Limited labeled data for diverse admission scenarios.
  • Ensuring model interpretability for admissions decisions.
• Solutions:
  • Augmented labeled data through synthetic data generation.
  • Utilized interpretable machine learning algorithms.
• Conclusion:
  • The model provided valuable insights into admission criteria, aiding in the selection process.

8. Volatility Forecasting in India

• Description:
  • Implemented time series analysis to forecast market volatility in the Indian financial market.
• Challenges:
  • Handling high-frequency financial data for accurate volatility predictions.
  • Adapting to sudden market changes and external events.
• Solutions:
  • Applied advanced time series models and machine learning algorithms.
  • Implemented adaptive models to quickly respond to market changes.
• Conclusion:
  • The forecasting model contributed to informed investment decisions in dynamic market conditions.

9. Housing Price in Mexico

• Description:
  • Analyzed and predicted housing prices in Mexico using machine learning models.
• Challenges:
  • Dealing with incomplete and noisy real estate data.
  • Enhancing model accuracy for diverse housing markets.
• Solutions:
  • Conducted extensive data cleaning and imputation.
  • Employed ensemble learning techniques for improved accuracy.
• Conclusion:
  • The predictive model provided insights for real estate investment strategies in the Mexican market.

10. Housing Price in Buenos Aires

• Description:
  • Developed a machine learning model to analyze and predict housing prices in Buenos Aires.
• Challenges:
  • Addressing cultural and economic factors influencing housing markets.
  • Enhancing model interpretability for stakeholders.
• Solutions:
  • Incorporated cultural and economic features into the model.
  • Used interpretable models and visualization techniques.
• Conclusion:
  • The model contributed to informed decision-making in the dynamic Buenos Aires real estate market.

11. Air Quality in Nairobi

• Description:
  • Implemented a machine learning model to predict air quality in Nairobi based on various environmental factors.
• Challenges:
  • Handling missing data in environmental sensor readings.
  • Ensuring real-time predictions for effective air quality monitoring.
• Solutions:
  • Applied data imputation techniques for missing sensor readings.
  • Utilized streaming machine learning for real-time predictions.
• Conclusion:
  • The model facilitated proactive measures for managing air quality in Nairobi.

12. Earthquake Damage in Nepal

• Description:
  • Developed a machine learning model to assess and predict earthquake damage in Nepal.
• Challenges:
  • Limited labeled data for earthquake damage scenarios.
  • Incorporating geological features into the model.
• Solutions:
  • Applied transfer learning techniques from related domains.
  • Included geological and structural features in the model.
• Conclusion:
  • The model provided valuable insights for earthquake preparedness and risk assessment in Nepal.

13. Bankruptcy in Poland

• Description:
  • Analyzed and predicted bankruptcy risk for companies in Poland using machine learning.
• Challenges:
  • Handling imbalanced data for bankruptcy prediction.
  • Ensuring model robustness in a changing economic landscape.
• Solutions:
  • Utilized ensemble learning and sampling techniques for imbalanced data.
  • Implemented model retraining strategies for adaptability.
• Conclusion:
  • The model provided early warning indicators for financial risk in the Polish business landscape.

Conclusion

These machine learning projects showcase a diverse range of applications and challenges. The solutions implemented demonstrate the ability to address complex problems and provide valuable insights. For source code and more details, please visit the GitHub Repository

Let’s Connect

• GitHub
• LinkedIn
• Email: [email protected]