Work

My professional experience and career journey. Here's an overview of the places I've worked and the roles I've held.

  • Upwork
    Sep 2023 - Current
    Data Scientist
    • Developed a real-time recommendation system using Kafka, Faiss, and PyTorch, improving customer engagement by 30% for an e-commerce client.
    • Built an end-to-end ML pipeline using Feast (feature store), MLflow, and Kubeflow, reducing model deployment time by 40%.
    • Designed a RAG-based AI chatbot for a finance client, integrating LangChain and vector databases, enhancing query response accuracy by 20%
    • Fine-tuned LLaMA 2 for a legal-tech startup, achieving 15% higher accuracy in case-law retrieval compared to GPT-4-turbo.
    • Engineered an automated supply chain forecasting model using transformers (TFT), reducing inventory costs by 25%.
    • Implemented self-supervised learning for anomaly detection in fraud analytics, cutting false positives by 35%.
  • Mu Sigma Inc.
    Sep 2020 - Sep 2022
    Senior Data Scientist
    • Led the development of a real-time data pipeline handling 1,000+ API calls daily, automating ingestion with PySpark, Apache Hive, and Airflow on AWS.
    • Built ARIMA-LSTM ensemble models for price and demand forecasting, achieving 87% accuracy and deploying scalable inference via Docker & Kubernetes.
    • Designed an anomaly detection system using Isolation Forests & Autoencoders, reducing anomalies by 65% with AWS Lambda for real-time fraud detection.
    • Developed a feature engineering framework using wavelet transforms & Fourier analysis, improving forecast accuracy by 17%.
    • Created Tableau dashboards integrating Flask APIs & AWS Redshift for real-time insights, optimizing query performance.
    • Built a machine learning pipeline to diagnose supply chain disruptions, reducing shortages by 38% and preventing $1.5M daily losses.
    • Developed a Bayesian MMM model, optimizing marketing spend and increasing ROI by 16%.
    • Designed a reinforcement learning algorithm for dynamic inventory management, cutting stockouts by 25%.
    • Built an RShiny simulator for MMM, boosting incremental revenue by 16%.