I build high-impact data solutions across computer vision, time-series forecasting, NLP, and more. Below are a few highlights:
-
Image Classification Project
-
Image Classification Project Transfer-learned MobileNetV2 in Colab to classify ten custom image categories—reaching 91% validation accuracy in under 20 epochs, with real-time training visualization and experiment tracking via Weights & Biases.
-
Time-Series Forecasting
-
Time Series Analysis & Forecasting Engineered windowed LSTM models to predict next-day rainfall across multiple stations—achieving a 0.48″ RMSE and R² = 0.86 on held-out data, outperforming persistence baselines by over 20%.
-
SQL Interpolation & Outlier Detection
-
Interpolation_Outlier_Detector_SQL Designed a three-stage PostgreSQL pipeline using advanced window functions and CTEs to auto-fill 100% of sub-hour gaps and flag critical ±3σ anomalies, streamlining data integrity workflows for real-time sensor networks.
-
Stock News Sentiment Analysis
-
Stock news Sentiment Analysis Built an end-to-end FinBERT pipeline that scrapes financial headlines, quantifies sentiment, and backtests trading signals—generating 12.5% annualized returns vs. 6.3% buy-and-hold, with automated report generation. STILL IN PROGRESS
-
Regression Exploration
-
Regression Exploration Conducted a full EDA and regression study using Linear, Ridge, and Lasso models on real-world data—engineering polynomial features to hit R² = 0.78 and RMSE = 2.4, complete with interactive visualizations.
-
Community Air Sampler Data & 6PPD-Q Dashboard
-
Community Air Sampler Data & 6PPD-Q Dashboard ELT lands field forms, GPS traces, and lab CSVs into a validated warehouse with geospatial and weather joins. Hotspot models and an auto refreshed public dashboard deliver school level exposure summaries, with QC logs linking each barcoded filter to time and place.