Almudena Zhou Ramírez López
Projects
Filter by technical skills
Note: You can select more than one with ctrl + click
-
-
Azure ML Fine Food Reviews: NLP-Based Sentiment Prediction Pipeline
A scalable NLP solution for analyzing customer reviews, designed to improve satisfaction and product offerings. The project leverages NLP, MLOps, and Azure ML with a scalable and reusable architecture.
Highlights:
- Sentiment Prediction Model: Built an adaptable model to classify customer sentiment across text formats, including reviews, tweets, and emails.
- Azure ML Pipelines: Developed automated, modular workflows for training and inference, ensuring adaptability across datasets.
- Batch Inference: Deployed scalable pipelines to handle large-scale sentiment analysis.
- Data Insights: Created notebooks to extract trends, satisfaction metrics, and behavioral patterns for actionable business insights.
- Scalable Framework: Implemented a Dockerized architecture with automated retraining capabilities.
Impact: Empowered data-driven business decisions by building production-grade machine learning systems on cloud infrastructure to transform unstructured feedback into actionable insights.
Tools: Python, Git, MLflow, Azure (Azure Machine Learning, Azure Blob Storage), Docker, sklearn, pandas, numpy, venv
-
Computer & Data Science Interview Assistant
A robust RAG-based chatbot designed to provide accurate answers to complex computer science and data science questions, utilizing a knowledge-based question-answering approach. The system includes real-time evaluation and monitoring, ensuring high accuracy, relevance, and performance.
Highlights:
- Retrieval-Augmented Generation System: Developed using Azure OpenAI, Ollama, Transformers, and ElasticSearch for embedding, retrieval, and answer generation.
- Data Sources and Preprocessing: Unified a Kaggle Q&A dataset and an interview question book into a clean, indexed knowledge base with thorough data preprocessing and structure checks.
- User Interface: Built a Streamlit app with a feedback system for users to rate response quality.
- Performance Monitoring: Integrated Grafana and PostgreSQL to track costs, response times, and feedback metrics in real time.
- Evaluation Metrics: Used MRR, hit rate, and RAG evaluation with GPT-4/GPT-3.5-turbo to ensure high answer accuracy and relevance.
- Containerized Deployment: Streamlined with Docker and Docker Compose for consistent and reliable application deployment.
Impact: Enhanced learning and interview preparation by providing an intelligent, real-time assistant capable of answering complex questions with precision and minimal hallucinations, while continuously monitoring and optimizing performance metrics to deliver reliable and relevant results.
Tools: Azure OpenAI, Ollama, SentenceTransformers, ElasticSearch, Grafana and PostGres DB,Streamlit, Python, Docker and docker-compose or podman, Conda, Git and Github
-
Student Performance: MLOps Pipeline
A comprehensive MLOps pipeline for predicting student grades, exemplifying a scalable and effective approach to deploying and maintaining machine learning models in production.
Highlights:
- Data Preprocessing and Model Training: Utilized Mage orchestrator for robust data handling and model development.
- Experiment Tracking and Model Registry: Implemented MLflow to streamline experiment tracking and manage models, with artifacts securely stored in AWS S3.
- Inference Pipeline: Enabled real-time model serving using AWS Lambda and Kinesis for streaming predictions.
- Monitoring: Integrated Evidently and Grafana to monitor model performance and detect data drift, with alerts sent via SES for proactive response.
- Infrastructure Management: Leveraged Terraform to ensure consistent and scalable infrastructure provisioning.
- Development Practices: Ensured high code quality with pre-commit hooks, automated workflows using a Makefile, and adherence to best practices.
- CI/CD Pipeline: Automated testing and deployment processes to ensure seamless and reliable updates.
Impact: Demonstrated the end-to-end development of a production-ready MLOps pipeline, enabling accurate and scalable predictions while ensuring reliability, maintainability, and efficient infrastructure management.
Tools: Python, Shell Script, Git, GitHub Actions, MLflow, AWS (Lambda with ECR, Kinesis, S3, IAM, CloudWatch), Terraform, Mage, Docker, docker-compose, Evidently, Grafana, LocalStack, Makefile, black, pytest, isort, pylint, sklearn, pandas, numpy, pyproject.toml, Pre-commit, venv
-
GitHub Archive: ETL Pipeline
A robust data pipeline to analyze GitHub’s top contributors and activity types, providing actionable insights into user behavior on the platform through data engineering skills.
Highlights:
- Data Extraction and Transformation: Utilized Mage as an ETL orchestrator and PySpark for data structuring, with structured data ingested into BigQuery, organizing events by time and activity type.
- Infrastructure Automation: Automated deployment using Terraform on Google Cloud, integrating Cloud Run for Mage hosting, GCS as a data lake, and BigQuery as the data warehouse.
- Batch and Stream Processing: Supported historical batch data and near real-time streaming with an hourly trigger for continuous updates.
- Data Visualization: Created a detailed Looker Studio dashboard to display insights on GitHub contributors, event categories, and non-commit actions for in-depth analysis.
Impact: Enabled detailed analysis of GitHub user actions through a scalable and automated pipeline, supporting data-driven decisions for platform activity insights.
Tools: Python, Git, Pyspark/Apache Spark 3, Google Cloud(Cloud Run, Google Cloud Storage, Google BigQuery), Terraform, Mage Orchestrator, venv, Looker Studio
-
Factories Pygame Boardgame: OOP and MVP Pattern
A refactored version of a strategy game I made, enhanced using Object-Oriented Programming (OOP) principles and the MVP (Model-View-Presenter) pattern to improve performance and flexibility for future updates.
Highlights:
- Enhanced Architecture: Leveraged OOP principles and the MVP pattern to restructure the game, ensuring clean separation of concerns and easier maintainability.
- Multilingual Support: Available in both Spanish and English.
- Game Modes: Supports multiplayer for two players or solo play against an AI opponent.
- Optimized Performance: Refactoring significantly improved game efficiency and responsiveness.
Impact: Revitalized a legacy project by applying a modular software development approach, improving maintainability and scalability.
Tools: Python, git, venv
-
-
-
Computer & Data Science Interview Assistant
A robust RAG-based chatbot designed to provide accurate answers to complex computer science and data science questions, utilizing a knowledge-based question-answering approach. The system includes real-time evaluation and monitoring, ensuring high accuracy, relevance, and performance.
Highlights:
- Retrieval-Augmented Generation System: Developed using Azure OpenAI, Ollama, Transformers, and ElasticSearch for embedding, retrieval, and answer generation.
- Data Sources and Preprocessing: Unified a Kaggle Q&A dataset and an interview question book into a clean, indexed knowledge base with thorough data preprocessing and structure checks.
- User Interface: Built a Streamlit app with a feedback system for users to rate response quality.
- Performance Monitoring: Integrated Grafana and PostgreSQL to track costs, response times, and feedback metrics in real time.
- Evaluation Metrics: Used MRR, hit rate, and RAG evaluation with GPT-4/GPT-3.5-turbo to ensure high answer accuracy and relevance.
- Containerized Deployment: Streamlined with Docker and Docker Compose for consistent and reliable application deployment.
Impact: Enhanced learning and interview preparation by providing an intelligent, real-time assistant capable of answering complex questions with precision and minimal hallucinations, while continuously monitoring and optimizing performance metrics to deliver reliable and relevant results.
Tools: Azure OpenAI, Ollama, SentenceTransformers, ElasticSearch, Grafana and PostGres DB,Streamlit, Python, Docker and docker-compose or podman, Conda, Git and Github
-
-
-
Azure ML Fine Food Reviews: NLP-Based Sentiment Prediction Pipeline
A scalable NLP solution for analyzing customer reviews, designed to improve satisfaction and product offerings. The project leverages NLP, MLOps, and Azure ML with a scalable and reusable architecture.
Highlights:
- Sentiment Prediction Model: Built an adaptable model to classify customer sentiment across text formats, including reviews, tweets, and emails.
- Azure ML Pipelines: Developed automated, modular workflows for training and inference, ensuring adaptability across datasets.
- Batch Inference: Deployed scalable pipelines to handle large-scale sentiment analysis.
- Data Insights: Created notebooks to extract trends, satisfaction metrics, and behavioral patterns for actionable business insights.
- Scalable Framework: Implemented a Dockerized architecture with automated retraining capabilities.
Impact: Empowered data-driven business decisions by building production-grade machine learning systems on cloud infrastructure to transform unstructured feedback into actionable insights.
Tools: Python, Git, MLflow, Azure (Azure Machine Learning, Azure Blob Storage), Docker, sklearn, pandas, numpy, venv
-
Student Performance: MLOps Pipeline
A comprehensive MLOps pipeline for predicting student grades, exemplifying a scalable and effective approach to deploying and maintaining machine learning models in production.
Highlights:
- Data Preprocessing and Model Training: Utilized Mage orchestrator for robust data handling and model development.
- Experiment Tracking and Model Registry: Implemented MLflow to streamline experiment tracking and manage models, with artifacts securely stored in AWS S3.
- Inference Pipeline: Enabled real-time model serving using AWS Lambda and Kinesis for streaming predictions.
- Monitoring: Integrated Evidently and Grafana to monitor model performance and detect data drift, with alerts sent via SES for proactive response.
- Infrastructure Management: Leveraged Terraform to ensure consistent and scalable infrastructure provisioning.
- Development Practices: Ensured high code quality with pre-commit hooks, automated workflows using a Makefile, and adherence to best practices.
- CI/CD Pipeline: Automated testing and deployment processes to ensure seamless and reliable updates.
Impact: Demonstrated the end-to-end development of a production-ready MLOps pipeline, enabling accurate and scalable predictions while ensuring reliability, maintainability, and efficient infrastructure management.
Tools: Python, Shell Script, Git, GitHub Actions, MLflow, AWS (Lambda with ECR, Kinesis, S3, IAM, CloudWatch), Terraform, Mage, Docker, docker-compose, Evidently, Grafana, LocalStack, Makefile, black, pytest, isort, pylint, sklearn, pandas, numpy, pyproject.toml, Pre-commit, venv
-
-
-
GitHub Archive: ETL Pipeline
A robust data pipeline to analyze GitHub’s top contributors and activity types, providing actionable insights into user behavior on the platform through data engineering skills.
Highlights:
- Data Extraction and Transformation: Utilized Mage as an ETL orchestrator and PySpark for data structuring, with structured data ingested into BigQuery, organizing events by time and activity type.
- Infrastructure Automation: Automated deployment using Terraform on Google Cloud, integrating Cloud Run for Mage hosting, GCS as a data lake, and BigQuery as the data warehouse.
- Batch and Stream Processing: Supported historical batch data and near real-time streaming with an hourly trigger for continuous updates.
- Data Visualization: Created a detailed Looker Studio dashboard to display insights on GitHub contributors, event categories, and non-commit actions for in-depth analysis.
Impact: Enabled detailed analysis of GitHub user actions through a scalable and automated pipeline, supporting data-driven decisions for platform activity insights.
Tools: Python, Git, Pyspark/Apache Spark 3, Google Cloud(Cloud Run, Google Cloud Storage, Google BigQuery), Terraform, Mage Orchestrator, venv, Looker Studio
-
Factories Pygame Boardgame: OOP and MVP Pattern
A refactored version of a strategy game I made, enhanced using Object-Oriented Programming (OOP) principles and the MVP (Model-View-Presenter) pattern to improve performance and flexibility for future updates.
Highlights:
- Enhanced Architecture: Leveraged OOP principles and the MVP pattern to restructure the game, ensuring clean separation of concerns and easier maintainability.
- Multilingual Support: Available in both Spanish and English.
- Game Modes: Supports multiplayer for two players or solo play against an AI opponent.
- Optimized Performance: Refactoring significantly improved game efficiency and responsiveness.
Impact: Revitalized a legacy project by applying a modular software development approach, improving maintainability and scalability.
Tools: Python, git, venv
-