↓ Hover any node - click to read
001

Predicting Mental Health from Wearable Data

Built an ML pipeline on biometric data from 17,500+ participants in the NIH All of Us dataset to analyze behavioral drivers of mental health. Identified sleep duration and daily step count as the strongest predictors, and used causal inference (LiNGAM) to uncover direct relationships between lifestyle patterns and depression risk.

Python K-Means LiNGAM Causal Inference NIH Dataset
MS Thesis
002

Agentic Research Assistant

A local research agent that searches, ranks, and synthesizes information into structured reports with citations and confidence scoring. Multi-step reasoning, tool use, end-to-end orchestration of retrieval and generation without external APIs.

LangChain Tavily ChromaDB Ollama
Building
003

Ad Copy Intelligence

Retrieval-and-generation system to analyze brand messaging and produce alternative ad copy with controlled emotional tone. Vector search retrieves similar campaigns; sentiment analysis assesses tone; LLMs generate variations that match or intentionally shift register.

LangChain ChromaDB VADER RAG
Building
004

Spotify Mood Engine

Analyzed personal Spotify listening data to cluster mood profiles and identify behavioral patterns over time. Treats music consumption as signal data, applying ML to understand how preferences shift across context and emotional state.

Spotify API K-Means scikit-learn
Building
005

MCP Server for IDE Productivity

Built an MCP (Model Context Protocol) server that connects Jira and Confluence into the IDE, so engineers can pull tickets, search docs, and update issues without leaving their editor. Reduced context-switching during development and made AI coding assistants ticket-aware.

MCP LLM Tooling Jira API Confluence API Developer Experience
Deployed
006

Guardrails for Test Case Generation

Designed and implemented a guardrails layer for an LLM-based test case generation system, validating outputs against schema, coverage, and safety constraints before they reached engineers. Improved trust and adoption of AI-generated tests in the development workflow.

LLM Safety Evaluation Python Production AI
Deployed