Reproduced key findings from the paper “Community-Based Fact-Checking on Twitter’s Birdwatch Platform” by replicating core data visualizations.

Built logistic regression models using 10K+ public data points to analyze predictors of helpfulness ratings and total votes on Community Notes, reporting standardized estimates with 99% confidence intervals.

Automated data collection and cleaning workflows using R and Unix Shell scripting to ensure full reproducibility.

Conducted an independent regression analysis to predict daily Citibike trip volumes using multiple linear regression with polynomial and interaction terms, applying k-fold cross-validation for model robustness.

Collaborated with NASA research teams to develop predictive models for harmful algal blooms (HABs) in freshwater lakes.

Applied atmospheric correction algorithms to enhance satellite-derived water quality data for CDOM analysis.

Built advanced machine learning models to estimate Dissolved Organic Carbon (DOC) using multi-sensor satellite data (Landsat 5/7/8, Sentinel-2) and 30+ years of in-situ measurements from 40+ Adirondack lakes.

Achieved strong predictive performance (R² = 0.90, MSE = 0.66 mg/L) using stacking regressors and neural networks.

Conducted trend analyses on water temperature across 130+ lakes, revealing long-term ecological shifts linked to climate and land-use change.

Built and deployed 3+ AI agents for automation tasks including text summarization, survey analytics, and documentation generation using Copilot Studio, reducing manual workload by 60%.

Developed a full-stack web application with React, Node.js, and a Flask-based API to process PDFs, DOCX, and images for automated workflow extraction and diagram generation using Gemini 2.0 Flash and Mermaid.js.

Integrated a Vue.js frontend for seamless file uploads and real-time interaction with the backend through secure, CORS-enabled endpoints.

Fine-tuned multiple lightweight Hugging Face transformer models (Gemma-2, Phi-2, TinyLLaMA, DeepSeek) using LoRA for financial question-answering tasks, achieving the highest numeric accuracy of ~25% with the best single model.

Evaluated zero-shot, context-augmented, and retrieval-augmented generation (RAG) strategies across Hugging Face and GUFF model families, identifying zero-shot RAG as the top performer for GUFF models with ~12% accuracy on numeric financial data.

Developed retrieval-augmented pipelines with LangChain and FAISS for efficient context chunking and similarity search, improving both answer relevance and semantic accuracy.

Delivered comprehensive technical support by diagnosing and resolving complex hardware, software, and network issues, minimizing downtime and ensuring business continuity for end-users.

Led the setup of IT infrastructure across 7+ locations, deploying and configuring over 300 laptops.

Collaborated with cross-functional teams and IT specialists to integrate and deploy technology solutions seamlessly, actively contributing to system improvements, and participating in strategic meetings to align on project goals and IT priorities.

Assisted in developing and maintaining detailed IT documentation, including network infrastructure diagrams, user manuals, and troubleshooting protocols, to streamline future support operations.

Collaborated with the KPMG Advisory team on analyzing 40,000+ NYC asthma cases, uncovering correlations and proposing partnerships with 3 external organizations.

Evaluated statistical and predictive models (OLS Linear Regression, Random Forest), achieving 87% accuracy in forecasting the impact of increased health insurance access on asthma prevalence in 25 Bronx neighborhoods.