Skills
AI & NLP / ML
Data Engineering & MLOps
Analytics & BI
Languages & Tools
Projects
AI Investment & Capital Deals Memo Copilot for Financial Analysis
AI system that ingests financial PDFs and auto-generates investment memos. Uses a RAG pipeline (LangChain + OpenAI + FAISS/Chroma) with AWS S3 storage and a Streamlit UI for extraction, document-aware Q&A, and memo drafting, featuring transparent agent execution. Designed to reduce analyst prep time and improve consistency.
End-to-End Data Pipeline: United States Non-Immigrant Visa Analysis and Prediction
Led a 4-member team to ingest, clean, and analyze U.S. non-immigrant visa data to study immigration patterns. Built an end-to-end pipeline with dbt + Python for transformation, feature engineering, and ML prediction.
Maximize Marketing: A Data Analysis Case Study
Analyzed fitness-app behavior; cleaned and joined data with SQL (CTEs, JOINs) and Excel (pivots, lookups). Communicated insights via Tableau dashboards and a GitHub write-up with recommendations.
SaaS Customer Churn Prediction, End to End MLOps
Built a production-ready churn prediction system for a SaaS dataset: clean/validate data, engineer features, train & compare models, and serve the best model via a FastAPI microservice. Workflow is fully reproducible (Docker + tests + CI/CD) with experiment tracking and basic drift/performance monitoring.
Figuring out Neural Networks: Classifying Breast Cancer from Mammography Images
Built a mammography classifier and compared a custom network to transfer-learning baselines (ResNet50, VGG16). Achieved 82% recall on the positive class using deep learning in Python.
Harry Potter and the Next Word: LSTM RNNs + Streamlit
Processed movie script lines (tokenization, n-grams, embeddings) and trained a 2-layer stacked LSTM to predict the next word. Improved accuracy by +15% (loss −32%) and deployed with Streamlit.
Summarize Anything: AI Content Compression - Summarization platform with Llama
Drop in long-form content and get an abstractive summary of the essential points in seconds, optimized for readability, powered by a Llama 3 based summarizer and a token-aware, cost-efficient orchestrator that handles long-form inputs without truncation.
Predicting with Volatility: Stacked LSTMs for AMZN Forecasting
Forecasted AMZN opening price with stacked LSTMs on Tiingo API data; achieved RMSE 7.4 on test and generated 30-day forecasts.