About Interests Work Experience Contact

UC San Diego, class of 2027

Raffi Dulgerian

Data science and mathematics undergraduate.

I am working toward a PhD in machine learning. My interests sit at the meeting point of applied mathematics and modern deep learning, from probabilistic modeling to building models that are efficient as well as accurate.

About

Background

I am a data science and mathematics double major at UC San Diego, graduating in 2027 with a 3.85 GPA. I am preparing to apply to PhD programs in computer science and applied mathematics, where I want to do machine learning research.

I am especially interested in understanding how models actually work, and how efficient they can be without losing what makes them useful. That interest lives right between the theory in my mathematics classes and the systems side of modern machine learning.

Alongside coursework I spent a year as an instructional assistant teaching mathematics to undergraduates, which sharpened how I explain and reason through ideas.

University of California, San Diego

BS, Data Science and Mathematics

Graduating2027
GPA3.85

Selected coursework

Machine LearningAlgorithmsOptimization ProbabilityData StructuresDatabases Principal Component Analysis

Interests

What I hope to work on

Machine learning research

Building deep learning systems that are efficient as well as accurate, including model compression and quantization for running large models on modest hardware.

Probabilistic modeling

Recovering hidden structure inside messy, real world data, and understanding what the learned states tell us about the system underneath.

The mathematics of learning

Optimization, geometry, and convergence. The theory that decides whether a model learns anything at all, and the reason I want to keep going into research.

Selected work

Projects

Math Reasoning Competition, CSE 151B

Spring 2026  ·  team competition

Built the inference pipeline for a deep learning course competition on mathematical reasoning. We ran a four billion parameter Qwen reasoning model in four bit precision on university GPUs, with no fine tuning, and added self consistency voting, type routed token budgets, and an answer repair pass. The submission reached 0.66 unified accuracy on the private leaderboard.

View repository
PythonPyTorchvLLMLLMsQuantization

The Greening Earth

2025  ·  data visualization

An interactive world map that visualizes decades of NASA MODIS satellite data to show where vegetation is growing or declining. Users compare a baseline year against a recent one to spot greening, drought stress, and deforestation across the globe.

View repository
JavaScriptD3Geospatial data

Hidden Markov Model for Electricity Demand

December 2025

Trained a three state hidden Markov model to sort hourly California electricity demand into low, medium, and high regimes from price and load data, using Baum Welch training and Viterbi decoding. Studied the learned transitions to measure how long demand persists in each regime.

PythonHMMsProbabilistic modeling

Loan Application Analysis

August 2025

Studied how lenders make decisions using real world loan data. Engineered financial and demographic features to estimate borrower risk and predict interest rates, and modeled disposable income from federal and state taxes to judge repayment capacity.

PythonPandasNumPyPlotly

N Gram Language Model

August 2025

Built an n gram language model with recursive back off trained on a text corpus, including the full pipeline for tokenization, counting, and probability normalization, validated with unit tests and doctests.

PythonPandasNumPyRegex

Experience

Where I have worked

Mathematics Instructional Assistant

UC San Diego

2025 to 2026  ·  La Jolla, CA

Ran office hours and discussion sections, breaking problems down to their core ideas and guiding students to reason through them together. Wrote grading rubrics with professors, graded homework and exams against them, and proctored exams while providing accommodations.

Office Clerk

Saddleback Appliances

2022 to 2024  ·  Lake Forest, CA

Managed appliance and plumbing inventory, organized stock by model number, and processed purchase orders through the store database. Handled customer calls and maintained the store website, including a clearance page to move older products.

Toolkit

Tools I reach for

Languages

  • Python
  • SQL
  • R
  • Java
  • JavaScript

Machine learning and data

  • PyTorch
  • scikit learn
  • Pandas
  • NumPy
  • vLLM

Environment

  • Git
  • VS Code
  • PyCharm
  • Eclipse

Contact

Happy to talk about research, machine learning, or PhD paths.