⚗️Chemical Engineering student at Universidad del Valle de Guatemala (UVG). I work on fermentation and fuel-pretreatment processes, and I use LLMs as working tools for the parts of research that eat the most hours: methodology drafting, data handling, and reporting.
Most of my work lives in a lab notebook and a LaTeX project. This profile is where the tooling side of it ends up.
🍾Bioethanol from acid whey — research group lead, 10 people Acid whey is a high-volume dairy byproduct with a disposal problem and a lactose content nobody is using. We run thermal deproteinization, enzymatic hydrolysis with A. oryzae β-galactosidase, and fermentation with S. cerevisiae. My focus is optimizing the variables across those stages. The theoretical ceiling for our feedstock is around 3% v/v ethanol, so the interesting question is not "does it ferment" — it's how much of that ceiling we can actually reach, and where the losses are.
⛽️Adsorbent pretreatment for biodiesel feedstocks — assistant lead, 12 people We evaluate adsorbents for pretreating waste oils, currently fish oil. The goal is to remove the need for acid esterification with H₂SO₄ — a step that is corrosive, generates acidic waste, and complicates downstream processing. My first academic output was a poster proposing a fixed-bed column methodology for this.
⚡️Caffeine recovery from coffee silverskin — team project Isolation and purification of caffeine from an aqueous silverskin extract: liquid-liquid extraction with food-grade solvents, recrystallization, and full mass balance. Coffee silverskin is a roasting byproduct, which matters in Guatemala more than most places.
Small tools for the unglamorous parts of lab work — gravimetric fermentation tracking, data cleanup, structured methodology drafting, and bibliography handling. I build these because my own projects need them, not as demos.
The first of these is revision-reporte, a Claude Code skill that runs the final check on my LaTeX lab reports: structure, APA 7 references verified against Crossref, SI units, and the internal consistency of experimental data.
The second is parque-vehicular-e10: an analysis of Guatemala's vehicle registry. 4,080,158 records the tax authority publishes as an undocumented flat file. built to check whether the E10 gasoline blend the country recently adopted is a problem for the national fleet. My first pass got the answer wrong: the aggregate distribution hid two opposite populations, and the real picture only appeared once I split motorcycles from light vehicles. The README walks through that mistake: it's a better account of what Claude Code is and isn't good for than any summary I could write here.
I work with Claude Code and API integrations. Earlier, in high school, I built a four-wheeled agricultural rover with a 7-in-1 NPK soil sensor that processed readings through an LLM API and returned irrigation and fertilization recommendations. It won two first places and one second place at inter-school science fairs and earned a recognition for contributions in AGTECH. That project is why I think LLM tooling belongs in engineering workflows and not just in software ones.
Tools: LaTeX / Overleaf · Zotero (biblatex, APA) · Python for lab data · Claude Code · API integrations
Languages: Spanish (native) · English (B2+)
Open to conversations about fermentation processes, biofuel pretreatment, or using AI tooling in experimental research — especially with people doing this work in Latin America.