Projects

Built a natural-language recommendation engine using a production-ready Retrieval-Augmented Generation pipeline.

• Orchestrated RAG with LangChain; embedded titles, genres, synopses, and metadata using OpenAI embeddings.
• Implemented LanceDB vector store for fast similarity search and top-N retrieval; optimized indexing for low-latency queries.
• Generated explainable recommendations with rationale and confidence ranking.

Designed an end-to-end ETL pipeline and analytics layer to enable reliable daily refresh SLAs and self-serve reporting.

• Automated ingestion from Google Cloud Storage to BigQuery using Apache Airflow and Python.
• Built dbt models with star schema design, data quality tests, and query optimization (partitioning, clustering).
• Delivered Power BI dashboards with DAX calculations for product and market analysis.

Real-time geospatial analytics platform for semiconductor fab monitoring (1st Place capstone).

• Tracked 2M+ RTLS devices; processed 100K+ logs/day with cleaning, geofencing, feature engineering, and time-series decomposition.
• Trained XGBoost Remaining Useful Life models achieving 82% accuracy; delivered event-driven alerts and KPI dashboards.
• Reduced downtime by eliminating manual monitoring overhead through real-time analytics.

PySpark analytics and predictive modeling for Los Angeles County listings with external feature enrichment.

• Built distributed pipelines with PySpark and Spark SQL; integrated crime and transit accessibility datasets.
• Achieved 76% accuracy in price forecasting through feature engineering and model optimization.

Computer vision system detecting mask compliance in surveillance videos for real-time violation monitoring.

• Used RetinaFace for face detection and Xception classifier for mask vs. no-mask inference.
• Processed video streams and surfaced violations for rapid response workflows.

Flask-based NLP chatbot for Covid-19 queries with deep learning intent classification.

• Implemented Keras Sequential model for intent classification with NLTK preprocessing.
• Reached 76% accuracy in question answering on benchmarked query sets.

Association rule mining for transaction datasets and personalized recommendations based on frequent itemsets.

• Applied Apriori and FP-Growth to discover item associations and frequent patterns.
• Built recommendation utilities leveraging discovered rules and customer behavior.