Hi! I'm currently a 4th year Data Science student at Northeastern University (5 year program, graduating 2027). I've worked on applied data science projects spanning time-series forecasting, biological data analysis, and model-driven applications. I really enjoy building and evaluating predictive models, designing data pipelines, and developing full-stack tools to explore and communicate results.

I’m excited to continue working with real-world, high-dimensional data and to contribute to projects where data science can support meaningful analysis and decisions!

Route-level time-series forecasting using U.S. flight data (2019–2025) and ERA5 weather reanalysis (Open-Meteo API).

• Evaluated XGBoost, LightGBM, LSTM, and TCN using 4-fold walk-forward validation
• Best performance achieved by XGBoost / LightGBM (11.25 min MAE, ~77.7% hit rate)
• Removing weather features increased error by 10.3%
• Built an interactive React + D3 dashboard for exploring forecasts, residuals, model comparisons, and performance diagnostics

Single-nucleus RNA sequencing analysis of the mouse Anterior Olfactory Nucleus integrating 10x Genomics data with the Allen Brain Cell Atlas.

• Applied PCA-based dimensionality reduction and clustering to analyze high-dimensional gene expression data
• Analyzed 36,863 nuclei, retaining 8,885 high-quality cells across 28 clusters
• Identified spatial segregation of excitatory neurons into dorsolateral and ventromedial populations
• Discovered Abi3bp as the top marker of contralaterally-projecting neurons using FDR-corrected testing

Document-store modeling and aggregation pipeline design using large-scale biomedical event data from the OpenFDA API.

• Ingested 5,000 nested FDA adverse event reports with hierarchical patient, drug, and reaction arrays
• Modeled pharmacovigilance data in MongoDB
• Computed death rates by administration route and demographic breakdowns using aggregation pipelines
• Implemented a PyMongo query interface returning JSON-serializable outputs

Systems comparison of PostgreSQL and Redis for Twitter-style workloads.

• Benchmarked approximately ~1M tweets and ~10K follow relationships
• PostgreSQL achieved higher write throughput; Redis provided faster timeline reads
• Explored architectural tradeoffs in fan-out strategies

Full-stack personal finance application with an ML-assisted receipt categorization pipeline.

• Flask API, React frontend, and MySQL backend
• Implemented a multi-stage categorization system with confidence-based routing
• Supports on-demand model retraining from user-corrected labels