Siddh Mandirwala

Developed a modern, responsive portfolio website using Next.js, React, and Tailwind CSS with smooth animations and theme support.

Integrated an AI-powered personal assistant trained to answer questions strictly about the developer's experience and projects.

Implemented secure lead capture, real-time streaming responses, and a polished user experience across devices.

Built a large-scale analytics and machine learning pipeline to analyze NYC MTA subway operations using 100GB+ of ridership, service, and accessibility data.

Processed data for 472+ subway stations with Apache Spark and PySpark to uncover insights on ridership patterns, system reliability, accessibility gaps, and post-pandemic recovery.

Developed a station-level ridership forecasting model using XGBoost with SHAP-based interpretability to support data-driven transit planning decisions.

Created an interactive data visualization platform analyzing spatio-temporal crime patterns in New York City over the past decade.

Designed narrative-driven visualizations to explore crime trends across time, geography, and offense types.

Enabled multi-dimensional filtering and interactive dashboards to transform complex public safety data into actionable insights.

Built a machine learning system to predict 30-day hospital readmission risk using real-world clinical data from the MIMIC-III and UCI Diabetes datasets.

Engineered demographic, clinical, and temporal features and evaluated Logistic Regression, Random Forest, and XGBoost models with metrics robust to class imbalance.

Demonstrated how predictive analytics can support proactive care planning, optimized resource allocation, and improved patient outcomes.

Developed an end-to-end Retrieval-Augmented Generation (RAG) system that answers domain-specific questions using data aggregated from GitHub, Medium, and LinkedIn.

Designed a modular pipeline for data crawling, cleaning, semantic embedding, and vector retrieval using Qdrant, with a fine-tuned GPT-2 model for answer generation.

Exposed the system through a REST API and an interactive Gradio interface, emphasizing efficiency, modularity, and practical LLM deployment constraints.

Analyzed the relationship between age, financial literacy, and saving behavior using a Bayesian causal modeling framework.

Modeled financial literacy as a mediating variable with probabilistic regression in PyMC, supported by causal graphs and posterior inference.

Applied posterior predictive checks and diagnostics to produce interpretable, uncertainty-aware insights into financial behavior.

Built a full-stack Instagram post scheduling platform with support for automated publishing and background job processing.

Implemented asynchronous task orchestration using BullMQ and Redis, backed by PostgreSQL for reliable data persistence and Docker-based deployment.

Designed the system to be scalable, modular, and production-ready for real-world social media workflows.