10/24/2025 updated
100 % available
AI/ML Engineer & Machine Learning Specialist
Angeles City, Philippines Bachelor of Science in Computer Science
About me
ML Engineer with 4+ years of experience building, fine-tuning, and deploying AI systems using LLMs, NLP, and embeddings. Skilled in Python, TensorFlow, scikit-learn, Node.js, AWS, Docker, and LangChain, with a track record of delivering scalable, production-grade ML solutions.
JavaScript (Programming Language)A/B TestingApplication Programming Interfaces (APIs)Artificial IntelligenceAirflowAmazon Web ServicesAmazon S3Data AnalysisAutomationC++ (Programming Language)CalibrationCloud ComputingCode ReviewContinuous IntegrationData CleansingExtract Transform Load (ETL)Data StoreDjango Web FrameworkExperimentationForecastingFraud Prevention and DetectionGithubR (Programming Language)Python (Programming Language)PostgreSQLMachine LearningMongoDBMySQLNode.JsPredictive AnalyticsQuery OptimizationTensorflowWriting DocumentationReproducibilitySciPySemanticsSQL DatabasesTableau (Software)Time SeriesTypeScriptWorkflowsFeature EngineeringPackaging and Processing DutiesPytorchFlask (Web Framework)Large Language ModelsPrompt EngineeringConvolutional Neural NetworksFastapiPandasMatplotlibScikit LearnKubernetesCoaching and MentoringPlotlyMachine Learning OperationsSafety PrinciplesAnomaly DetectionRecurrent Neural NetworksDockerMicroservices
- Core — Python, TypeScript/JavaScript (Node.js), R, SQL, C++; clean, modular code; CI/CD.
- AI/ML — TensorFlow, PyTorch, scikit-learn, Pandas/SciPy; feature engineering; model selection and tuning; evaluation (ROC-AUC, precision/recall).
- LLMs & NLP — OpenAI/LLaMA/Gemini, embeddings, semantic search, prompt engineering, RAG, vector DBs (Pinecone, FAISS, Chroma).
- Time-Series & Predictive — LSTM/GRU/CNN, ARIMA, Prophet; forecasting, anomaly detection, calibration, A/B testing.
- MLOps — Docker, Kubernetes, GitHub Actions; experiment tracking; model packaging/serving; reproducible pipelines.
- Data Eng — ETL/ELT, data cleaning, Airflow; scalable batch/real-time pipelines.
- APIs & Services — FastAPI, Flask, Django/DRF, Node.js microservices; REST design, auth, observability.
- Cloud — AWS (Lambda, S3, containerized deployments); cost/perf optimization.
- Data stores — PostgreSQL, MySQL, MongoDB, S3; schema design and query optimization.
- Analytics & Viz — Tableau, Plotly, Matplotlib; automated reporting workflows.
- Practices — TDD, code reviews, documentation, mentoring; security and compliance mindset.
- Impact Highlights — Fraud detection latency ↓50%; campaign response ↑20%; retraining time ↓40%; loan approval accuracy ↑18%.
Languages
EnglishFluent
Project history
- Designed and deployed LLM-powered fraud detection and customer segmentation systems, reducing fraud detection latency by 50% and increasing campaign response rates by 20%.
- Implemented semantic search and embeddings for financial risk modeling using Pinecone and FAISS.
- Integrated ML models into Node.js + FastAPI microservices, enabling seamless use across distributed teams.
- Developed data pipelines with Apache Airflow, cutting retraining time by 40%, and containerized models with Docker for scalable deployments.
- Collaborated with DevOps to orchestrate deployments on Kubernetes, ensuring high availability and fault tolerance.
- Mentored junior engineers, reviewed code, and introduced best practices for prompt engineering and model evaluation.
- Built and deployed credit risk scoring models (XGBoost, logistic regression), improving loan approval accuracy by 18%.
- Conducted NLP-based fraud detection using transaction pattern analysis, reducing fraud-related losses by 15%.
- Automated reporting pipelines in Python and Tableau, improving efficiency by 40%.
- Developed time-series forecasting models (Prophet, ARIMA) to predict loan demand with <7% error.
- Ensured data security and compliance with BSP guidelines, applying MLOps practices for production deployment.
- Assisted in cleaning, preprocessing, and structuring financial datasets for machine learning applications.
- Prototyped an early-warning loan default detection system, supporting the transition to a production-grade risk scoring tool.
- Conducted exploratory data analysis (EDA) to identify correlations between repayment behaviors and customer demographics.
- Helped senior data scientists evaluate multiple ML models using metrics like ROC-AUC, precision, and recall.
- Documented data workflows, schema designs, and preprocessing scripts, improving knowledge transfer and onboarding.
- Automated ETL scripts for financial datasets, reducing manual data preparation by several hours weekly.
- Contributed to team sprint meetings and provided support for ad hoc analysis tasks requested by managers.
Certificates
Kaggle Certifications
Kaggle2025
PACUCOA Accreditation
Angeles University Foundation2021
Portfolio
Credit Risk Scoring Application
- Focus: User interface and immediate result.
- Description: A clean UI mock-up of a loan application tool, showing an applicant's details and the system-generated "Risk Score" (e.g., "780 – Low Risk") with an "Approved" status.
Credit Risk Model Evaluation (ROC Curve)
- Focus: Model performance.
- Description: An ROC-AUC curve with a clear AUC score highlighted, along with precision, recall, and F1-score metrics, demonstrating model accuracy.
Credit Risk Factors/Explainability
- Focus: Model transparency and key drivers.
- Description: A waterfall chart or bar chart illustrating the positive and negative contributions of different factors (e.g., Credit Score, Debt-to-Income, Employment History) to a loan applicant's risk score.
The “Anomaly Alert” Dashboard
- Focus: Direct output and impact.
- Description: A dashboard screenshot showing a transaction flagged as suspicious, alongside KPIs demonstrating reduced latency and improved detection rates.
Fraud Detection Model Architecture (LLM-Enhanced)
- Focus: Technical complexity and methodology.
- Description: An architectural diagram illustrating the flow of raw data through LLM analysis, anomaly detection, vector search, and final alert.
Feature Importance for Fraud Detection
- Focus: Model interpretability.
- Description: A bar chart showing the top N features that contribute most to the fraud detection model's decisions (e.g., Transaction Amount, Location, Frequency, Time of Day).
Customer Segmentation Dashboard
- Focus: Visual representation of customer groups.
- Description: A dashboard showing a scatter plot with distinct, color-coded customer clusters, along with key characteristics for each segment and a KPI for campaign response rate.
Word Cloud/Text Analysis of Customer Feedback
- Focus: Qualitative data and deeper insights.
- Description: A visualization showing three separate, small word clouds, one for each major segment, displaying common words/phrases from their feedback or support tickets to help tailor communication.
Time-Series Forecast (Actual vs. Predicted)
- Focus: Prediction accuracy and confidence.
- Description: A clean line graph showing the historical data, the forecasted line, and the 95% confidence interval band, highlighting the achieved <7% error margin.
Model Comparison: ARIMA vs. Prophet vs. LSTM
- Focus: Methodology and model selection.
- Description: A simple bar chart comparing the performance (MAE or MAPE) of different time-series models (ARIMA, Prophet, LSTM) that were tested, justifying the model choice.
Dashboard: Predictive vs. Actual Cash Flow Impact
- Focus: Decision-making support.
- Description: A dashboard showing the immediate operational impact—e.g., predicted Cash Flow Requirement vs. Actual Cash Flow, and a KPI showing improved resource allocation efficiency.