Projects
A few projects I'm proud of, and a few more I learned a lot from.
Deep Dives
Auto iMessage ingestion, calendar sync, project tracking, AI todos, and Garmin biometrics in a Monet pixel art UI.
- • Auto iMessage ingestion pipeline that parses conversations into structured action items, reminders, and context for AI-driven task generation
- • Calendar + project integration that syncs events, deadlines, and repo activity into a unified daily command view
Gemini-powered French reader with scene segmentation, translation, vocab, and quizzes.
- • Six-stage pipeline: PDF extraction, heading stripping, sentence segmentation, Cloud Translation, chunking, and Gemini annotation
- • Context-aware scene segmentation that passes previous scene summaries to maintain narrative coherence across chunks
Two-stage PyTorch CV pipeline: hand detection then finger-count classification.
- • Two CNNs (~7.5M params each): 4-layer conv stack + 9-layer dense head on 28x28 grayscale inputs
- • Custom dataset with canvas-based bounding-box annotation tool and albumentations augmentation
Wildfire research pipeline: Earth Engine satellite downloads, GOFER-format JSON normalization, scikit-learn logistic regressions, and a canvas-based fire viewer.
- • Earth Engine download and normalization pipeline
- • Locational spread + continuation regression workflows
CFR+ poker solver: 100K iterations of external sampling MCCFR, card bucketing, and a web UI with live range heatmaps.
- • CFR+ with external sampling: 100K iterations, regret clamping, bucket-cached traversal across all 4 streets
- • K-means preflop abstraction (10 buckets) + empirical CDF postflop bucketing with sampled board completions
Project Archive
Prediction market maker: real-time sentiment-driven spread pricing, inventory risk management, multi-venue quote execution, and a React monitoring dashboard.
Statically exported Next.js portfolio with live Garmin and Spotify data widgets, filterable project archive, and per-project case study pages.
Spring Boot commute solver: multi-leg trip matching across MBTA lines with walk/transfer buffers and a leave-by planner UI.
Hex-grid Reversi: 48-class MVC architecture with cube coordinates, composable AI strategies via TryTwo combinator, and adapter-pattern provider integration.
Agentic study pipeline: CrewAI agents + ChromaDB RAG turn PDFs into Anki flashcard decks and difficulty-weighted LaTeX cheat sheets.
Custom C shell in 793 lines: recursive arbitrary-depth piping, input/output redirection, tokenizer with quote handling, and 5 built-in commands.
N-Queens GA solver with tournament selection (k=3), single-point crossover, and 1/N adaptive mutation rate—solves N=8 in ~500 generations.
