Project Overview
OLTP/OLAP warehouse, a FastAPI quote engine, and a 0–100 viability score blending LCOE, payback, NPV, solar resource, and local market maturity.
About
I'm a data engineer in the Bay Area, currently in the M.S. in Applied Data Intelligence at San José State University. I build warehouses, pipelines, and small tools that turn messy public and operational data into something you can actually run.
The longer pages on this site are research notes I care about. They are optional depth, not the pitch.
Projects
☀️ Solar Viability Analysis
Address-to-financial-analysis for California rooftop solar: a 2.6M-row DuckDB warehouse, a 25-year cashflow model, and live NREL PVWatts + utility-rate detection across PG&E, SCE, SDG&E, and municipal utilities. Built for DATA 201 at SJSU.
City Reports
Deep-dive reports for the 15 largest CA cities plus Stanford & Berkeley: 25-year cashflows, payback curves, and utility-specific NEM 3.0 economics.
ZIP Heatmaps
Interactive viability maps across all 2,593 California ZIP codes, with confidence tiers drawn from real Berkeley Lab installation data.
AI & Machine Learning
Multi-Armed Bandit Agents
Exploring reinforcement learning through the classic multi-armed bandit problem.
The Speedy, Q-learning Crawler
A year of RL tuning reached only 42% of optimal, then a 2002 model-based algorithm (R-MAX) closed the gap to 100%.
From Counts to Context
HMM → LSTM → BERT for NER. A leakage bug in the validation split, then DistilBERT at test F1 0.888. The older Viterbi writeup is linked from the page.
The Value of Knowing You
A topological curriculum from Bayes nets to counterfactuals: an ad agent climbing Pearl's ladder. Three interactive widgets and a modernized, runnable decision network.
Interactive Visualizations
Self-contained, in-browser explorables — no backend, rendered live on canvas.
City Cartography
A topographic recreation of anvaka/city-roads: OpenStreetMap road networks drawn live, layered over real elevation contours, hillshade, water, and green space.
Mandelbrot Explorer
A deep-zoom fractal renderer: web-worker Mariani–Silver subdivision with Hilbert-curve ordering and smooth iteration coloring, down to the precision floor.
The Shape of People to Come
An econometric atlas of post-industrial demographics: life-cycle capital structure, a six-instrument panel beyond birth rates, and 236 countries scored on an adjustable demographic-strategic index (UN WPP 2024).
Knowledge Source Scrapers
Three complementary Python scrapers forming the raw-sources layer of an LLM knowledge pipeline for cognitive science.
Wikipedia Scraper
Bounded BFS crawl with canonical page keys, SQLite graph store, and a 6-point soup validator. 834 outlinks from a single seed article.
SEP Scraper
Stanford Encyclopedia of Philosophy: peer-reviewed concept backbone with hand-curated cross-references and a built-in query layer.
arXiv Scraper
Atom API metadata ingest with category-driven queries, PDF download, and text extraction across 35 CS subcategories.
Econometrics
Does Bitcoin Follow a Power Law?
An interactive econometric autopsy. Eight lenses — HAC inference, residual stationarity, AIC model selection, OU mean-reversion, LPPLS bubbles, structural breaks — scrutinize the power-law claim with Plotly, statsmodels & seaborn on a decade of data.
Resume
Contact
Email, LinkedIn, or GitHub.