This site!
Portfolio
From-scratch vibe-coded portfolio with HTML & CSS.
Stack: GitHub Pages · Claude Code · HTML/CSS
Fig. 03.a — Meta-analysis
Asparagopsis*
📄 Citation: Goloja, C., Povejsil, N., Roque, B.M. et al. A meta-analysis establishes bromoform dose and forage-based models to evaluate the antimethanogenic effects of Asparagopsis spp. feed additive. BMC Agric. 2, 4 (2026). https://doi.org/10.1186/s44399-025-00030-w
Asparagopsis is a red seaweed that, when fed to cattle in small doses, dramatically reduces their methane emissions. This paper consolidates results from dozens of individual trials into dose-response and forage-adjusted models — work that directly supports regulatory approval pathways and carbon credit methodologies for the livestock industry.
Conducted during my time at Symbrosia, where I also contributed ROI analyses used in investor pitches and modeled emissions reductions for commercial deployment scenarios.
Fig. 02.a — Pipeline architecture
HydroScale*
🏆 5th Year MIDS Capstone Award
Forecasting Water Usage Efficiency in Data Centers. American data centers collectively consume about 300,000 gallons of water per day — roughly equal to 100,000 homes — and more than 20% withdraw water from medium to high stress watersheds. With AI driving a projected CAGR of 22% in global DC capacity (generative AI workloads alone: 39%), water requirements will strictly increase.
We developed HydroScale, a platform providing granular 72-hour ahead WUE forecasts across the United States. Leveraging weather, energy, and operational data, it delivers on-site forecasts tailored to individual data centers and off-site forecasts considering broader regional factors — geospatial estimates of where water usage will be most efficient, enabling geographic load balancing and job scheduling.
Tags: Python · Data Engineering · Machine Learning · GIS · Transfer Learning
Fig. 01.a — Water Reuse Potential Index, U.S. water service boundaries
GeoNdxR*
Open-source GIS tool for environmental indices. Water reuse — defined by the EPA as "the practice of reclaiming water from a variety of sources, treating it, and reusing it for beneficial purposes" — is too often seen only as a solution for water-scarce areas, rather than for its full potential to lower flood risk, reduce sewer overflows, and minimize nutrient discharge.
geondxr is a low-code R tool that uses WebR to run index re-calculation and interactive mapping directly in the user's browser — no server needed, free to host on GitHub Pages. Index creators preprocess data, calculate the index, and ship it as a single .html file.
📝 Aug 2024: This project is based on The Water Research Foundation (WRF) project 5197, funded under Assistance Agreement No. 84046201 awarded by the U.S. Environmental Protection Agency (EPA) to WRF. The views expressed are solely those of the authors and do not necessarily reflect those of WRF or EPA.
Tags: R · GIS · Statistics · Web Dev · Visualization
§
Academic Final Projects
Fig. A7.a — Image augmentation and modeling pipeline
Wildfire Prediction
Since the 1980s, wildfires have intensified due to climate change, posing significant threats to ecosystems and economies. Early wildfire risk assessment is crucial for resource allocation, prevention, and disaster response. By analyzing aerial photos, ML models can assess wildfire risk to aid proactive measures like vegetation management.
Augmented the public dataset of Canadian pre-wildfire and no-wildfire labelled aerial images to build a classification system. Trained and compared several supervised deep learning models against classical methods on unseen test data.
Results: Top model — Gradient Boost Classifier — achieved ~96% test accuracy and F2 Beta Score of 0.97. All models exceeded 90% accuracy on this binary classification task.
Tags: Computer Vision · Python · Machine Learning · Remote Sensing · Transfer Learning
Fig. A6.a — BART system knowledge graph for pickup routing
Wrangling 3rd-Party Sales
End-to-end data engineering project for a subscription-based meal delivery service. Worked in AWS (VM + Docker cluster with Anaconda and Postgres containers). Transformed and analyzed data using Pandas and SQL. Created an interactive geographic heat map of customer locations and optimal storefront locations using Google Maps and Python.
Results: Identified optimal pick-up locations along BART lines · Dijkstra's Algorithm selected for delivery route optimization · Practical business cases outlined for MongoDB, Redis, and Neo4j.
Tags: Data Engineering · SQL · Python · GIS
Fig. A5.a — CNN architecture and validation accuracy
Detecting Cancer from Histopathological Images
Given vast numbers of histopathological images and diminishing medical staff worldwide, can neural networks reliably detect cancer and be integrated into radiologists' workflow? This project tests a basic CNN on a binary cancer/no-cancer classification task on labelled images.
Implemented a two-layer CNN (16 and 32 filters, kernel size 5) with two fully-connected layers in PyTorch. Optimized with SGD and cross-entropy loss over 50 epochs.
Minimizing false negatives is critical here — the stakes for a missed cancer diagnosis are higher than for a mistaken one. Further augmentation or architecture changes could push accuracy higher.
Tags: Machine Learning · Computer Vision · Python · Deep Learning
Fig. A4.a — Academic final paper
Algorithmic Governance & Predictive Policing
This report focuses on predictive policing — the practice of using algorithms to anticipate criminal activity before it occurs. Two case studies: U.S. company PredPol and Chinese company Megvii. Each illustrates the ethical issues, public reception, and government regulation involved.
The report highlights how rapidly developing technical approaches have outpaced government regulation, and explores the social implications of increasingly technocratic law enforcement. Written for regulators exploring how cases have unfolded in two countries with contrasting ethical and regulatory standards.
Tags: Research · Writing · Policy and Regulation · Governance · Machine Learning
Fig. A3.a — Academic final paper
Tropical Medicine & Malaria Control
Certain facets of human geography — migration, lack of social organisation, environmental degradation — are known to contribute to infectious disease prevalence. Yet Western public health instead looks towards medical intervention and disease-specific prevention to address the resulting epidemics after they have already spiralled out of control.
The prioritisation of downstream interventions stems in part from the imperial legacy of "tropical medicine," a medical speciality that originated as an arm of the British Empire whose purpose was to control diseases that inhibited expansion and productivity in their overseas extraction colonies. While today global health organisations tout the "human right to health," the economic metrics they use to quantify the potential impact of aid severely undermine the humanitarian sentiment.
This paper argues that the roots of the technology-based global health approach to malaria control today can be found in the British imperial legacy in Latin America and the Caribbean — a legacy that persists despite recent advances towards reducing the threat of malaria in the region.
Tags: Research · Writing · Public Health · Epidemiology · Policy and Regulation
Fig. A2.a — Academic final paper · Summary below
Methyl Mercury & Child Neurodevelopment
"The Effect of Methyl Mercury Poisoning Due to Fish Consumption on Neurodevelopment of Children."
First synthesized in a London laboratory in the 1860s, methyl mercury has remained over-abundant in the environment due to chemical manufacturing and power plant byproducts. Mercury bioaccumulates easily from microorganisms to large mammals. In populations that consume fish frequently, most daily mercury intake is in the methyl mercury form.
Fetal cells contain 30% more mercury than maternal cells in Swedish, Japanese, and Iraqi fishing populations. Children's incomplete blood-brain barrier allows lipophilic methyl mercury to settle in the central nervous system, causing disrupted neuronal cell division and migration during brain development.
This paper covers exposure assessment, toxicology, epidemiology (including the Minamata and Iraqi poisoning outbreaks), and risk management through the Minamata Convention, Clean Air Act, Clean Water Act, and EPA emission standards.
Tags: Public Health · Policy and Regulation · Writing · Research · Epidemiology
Fig. A1.a — Supply/demand and market structure models · Summary below
Land Use Permits in Napa Valley
The Winery Definition Ordinance of 1990 amended Napa County Code to establish the "75% rule" (at least 75% locally grown grapes), set minimum parcel size requirements for new wineries, and aim to "reduce densities and thereby lessen local visual, traffic, air, noise, and groundwater impacts and reduce the conversion of viable agricultural land."
In this analysis, I apply supply and demand elasticity models to illustrate the dynamic between agricultural businesses and other potential land buyers. Because the government restricts land access through permits, land is both excludable and rival. I also examine how the ordinance shifted the market from near-perfect competition toward an oligopolistic structure — raising prices, reducing output, and pushing out small independent producers.
One key limitation: the negative externality model fails to account for harms from agriculture itself (soil erosion, pesticide pollution). The ordinance addresses urban and industrial development impacts but overlooks those from agricultural production.
Tags: Policy and Regulation · Writing · Research