building intelligent systems that turn data into real decisions
I build end-to-end ML projects with a practical focus: clean pipelines, measurable evaluation, and results that hold up outside a notebook. I like problems where clarity matters as much as accuracy.
reliability, interpretability, and systems you can iterate on
models and tooling that turn data into decisions with minimal friction
Selected work across NLP, vision, forecasting, and analytics.
End-to-end fake vs real news classification with TF-IDF + Logistic Regression, evaluation charts, and a Streamlit app.
Fine-tuned BERT for tweet sentiment (negative/neutral/positive) with full evaluation: confusion matrix, ROC curves, and word clouds.
CNN encoder (ResNet-50) + LSTM decoder in PyTorch, with preprocessing, vocabulary building, BLEU evaluation, and inference examples.
LSTM forecasting on OHLCV data with sliding windows, early stopping, and metrics (RMSE/MAE/MAPE) plus clean prediction plots.
Data-driven site selection: SQL (SQLite) prep + regression modeling to estimate monthly profit, with feature importance and diagnostics.
Generate synthetic transactions, mine frequent itemsets (Apriori/FP-Growth), derive association rules, and export clean charts/CSVs.
Metrics, confusion matrices, and what I look for beyond accuracy.
Common leakage traps, windowing mistakes, and what to validate early.