Open to full-time roles from June 2027
I'm Michael, a Northeastern data science student building data and ML products, from an iOS app to sports sensor analytics.
I work where data, machine learning, and software meet, turning messy information into tools people actually use. I'm especially drawn to applied AI: embeddings, retrieval systems, and the quiet infrastructure that makes intelligent products feel effortless.
Track & field analytics for coaches. Insole sensors send data over Bluetooth to a web and iOS app, which computes ground contact time, flight time, step counts, and hurdle splits after each run.
App Store release, Sept 2025
Guitar ear-training app with a daily chord challenge, progressive hints, practice modes up to barre and blues chords, friends, and a global leaderboard. Built natively in Swift and SwiftUI.
Traces 50 years of ICSE research (~11,000 papers, 1976 to 2025) with one BERTopic model fit across all venues, grouping 174 topics into 10 themes.
PPO agent for Quoridor, warm-started with behavioral cloning and refined through self-play on a CNN board encoder. Beats a random bot about 100% of the time and a heuristic bot about 85% when moving first.
PyTorch, PPO, self-play, behavioral cloning
Student Experience Matching System: matches Northeastern students with experiences from their survey answers, using sentence-transformer embeddings and pgvector similarity search. Built in three weeks.
Django, React, Sentence Transformers, pgvector, AWS
Classifies 13 types of logical fallacies across ~3,800 labeled arguments, comparing TF-IDF, deep averaging networks, handcrafted argument features, and fine-tuned transformers. Class-weighted RoBERTa scored best (test macro-F1 0.45).
Python, PyTorch, RoBERTa, scikit-learn
Contributed the Sanity CMS setup, projects portfolio, leadership and consultant profiles, and testimonials to NExT Consulting's Astro site, deployed on Vercel.
Astro, Sanity CMS, Vercel
Answers questions from class-note PDFs with local LLMs through Ollama. Compared three embedding models across FAISS, Chroma, and Redis on ingestion and retrieval speed and memory.
Python, FAISS, Chroma, Redis, Ollama
Python toolkit that reads SEC 10-K filings and compares companies by word frequency, keyword sentiment, word clouds, and Sankey diagrams.
Python, NLTK, TextBlob, Plotly
Jun 2026 - Present
Boston Consulting Group (BCG)
Building tooling for AI knowledge pipelines: a tagging tool for a knowledge-database ingestion pipeline, reliability fixes to an agentic case-curation pipeline, and a rerun-safe validation job. Leading a study of whether AI-assisted development pays off across 6+ squads.
Jan 2026 - Present
Generate Product Development Studio
Directing the data branch across all project teams (35+ members), setting project strategy, data architecture, and workflows, and hiring 6 Technical Leads for Fall 2026. Promoted from Data Analyst, where I built StrideTrack's Bluetooth sensor connection (web and iOS) and run-recording flow.
Jan - May 2026
Khoury College of Computer Sciences
TA for Database Design (CS3200), supporting 200+ students with SQL, data modeling, normalization, and query optimization through office hours and grading.
Jul - Dec 2025
NExT Consulting
Led DevOps for a six-person team across three client projects. Architected Via Data, an AWS pipeline (ECS, Timestream, Grafana) for 1-second manufacturing IoT data, and built Google OAuth, surveys, and AWS deployment for SEMS, an embedding-based student matching app.
Jan - Jul 2025
Northeastern University
Built a licensure-tracking system for the Office of the Provost, monitoring state and territory licensure requirements for 51 programs via automated scraping and AI-powered analysis.
I'm always up for meaningful projects and new opportunities, whether in data, ML, or anything worth building well.