// नमस्ते

sulav@lexington:~$ whoami

Sulav Sapkota

Math and CS student at Washington and Lee, class of 2029. I build data and ML projects.

Resume GitHub LinkedIn Email

Lexington, VA

01 // projects

Projects

The wildfire charts and the earthquake explorer use real results from the repos. The gossip network is a small live version of my simulator. The microplastics and dashboard visuals are illustrations.

Real results from the repo. Random Forest in every row except the last, and the axis starts at 50%.

featured · research

Wildfire retardant drops: which side was protected?

A geospatial machine-learning pipeline that predicts which side of an aerial fire-retardant drop is the protected side, using day-of-burning rasters and the drop geometry. Research at Washington and Lee. The biggest jump came from comparing each drop's two sides across the whole fire, not just next to the drop.

  • 91.5%fire-grouped 5-fold CV (20,022 drop sides, 420 fires)
  • 89.6%on 2020–2021 drops labeled by different people
  • 68.9%for a simple rule baseline
  • Python
  • GeoPandas
  • Rasterio
  • XGBoost
  • scikit-learn

Code

Real data: shallow quakes (70 km or less) from the dataset. Drag the cutoff and watch b change.

Himalayan Quake Atlas

A Python and SQLite pipeline over 10,125 USGS earthquakes (1990–2026) with an interactive map and charts. I estimate the Gutenberg-Richter b-value, and the main lesson is that the answer swings from 0.79 to 1.27 depending on the completeness cutoff you pick, so the dashboard shows the whole curve and explains the choice.

  • Python
  • SQLite
  • pandas
  • Chart.js
  • Leaflet

Live demo Code

A small live version of the model: 110 nodes, 15% message loss. Pick a fanout and watch it spread. The real runs use 1,000 nodes in C++.

Rumor Mill

A C++20 discrete-event simulator of gossip spreading through a network, with a priority-queue engine, seeded randomness, message loss and latency. Across 180 runs on a 1,000-node graph, raising the fanout from 1 to 6 cut the time to reach 99% of nodes from 10.6 to 1.8, but total messages stayed around 10–11k, mostly duplicates.

  • C++20
  • Python
  • Make

Code

Illustration of the task, not project data.

Microplastics detection

Deep learning to find and classify microplastics in stereo-microscope images, at Kathmandu Institute of Applied Science. U-Net and VGG16 models reached 82% test accuracy on binary detection (783 images) and were extended to five plastic types: foam, pellet, film, fiber and fragment. YOLOv8 counted particles (0.40 mAP50). It won Best Poster at MoChWo 2024 in Kathmandu.

  • TensorFlow
  • YOLOv8
  • OpenCV

Research project, no public code

Illustration of the layout, not real data.

Health indicators dashboard

A React and Node.js dashboard that charts 12 health indicators across Nepal's 7 provinces, with a map you can click to zoom into individual districts. Built at Hello World I.T Solutions and live in production.

  • React
  • Node.js
  • Chart.js

Client work, no public code

02 // about

About

I'm from Kathmandu, and I now study math and computer science at Washington and Lee University in Virginia. I like problems where careful math meets messy real data: burn maps from wildfires, microscope images, and 36 years of earthquake records. I'm happiest when I can explain exactly why a number is what it is, including where it could be wrong.

Outside the lab, I co-founded Yukti Foundation, an empowerment-focused NGO in Kathmandu. Our first project installed a sanitary pad vending machine for women at Dashrath Stadium, paid for with the $280+ we raised, and we volunteer at community sports events like the SAFF Women's Tournament. I also shadow stock pitches at the Williams Investment Society, and before college I competed in chemistry and informatics olympiads and built a maze-solving robot with Arduino.

profile.json
{
  "from":      "Kathmandu, Nepal",
  "studying":  "B.S. Math & CS (intended), W&L '29",
  "codes_in":  ["Python", "C++", "JavaScript",
                "SQL", "R"],
  "speaks":    ["Nepali", "English", "Hindi"]
}

03 // experience

Experience

  1. June 2026 – present

    Machine Learning Research Assistant · Washington and Lee University

    Predicting the protected side of aerial fire-retardant drops from burn maps, with leakage-safe evaluation.

  2. June 2023 – June 2025

    Machine Learning Researcher · Kathmandu Institute of Applied Science

    Segmentation, classification and detection of microplastics in microscope images.

  3. June 2024 – August 2024

    Student Researcher · Water Stories of Patan

    Led a 6-member team on a digital water-heritage project: 30+ local surveys, policy mapping from 1963 to 2015, and a public website.

  4. January 2021 – June 2024

    Software Developer · Hello World I.T Solutions

    React, Node.js and Django projects, including a health dashboard in production and a fixtures site for a Nepali-owned soccer team in Australia.

04 // contact

Say Hello!

I'm looking for software engineering and machine learning internships. The quickest way to reach me is email.

ssapkota@mail.wlu.edu Resume (PDF) GitHub LinkedIn