Janelle Sampson — Machine Learning & Software Engineer

About

I work at the intersection of full-stack engineering and machine learning. I build the models, the pipelines that feed them, and the systems that put them in front of users. Lately that's meant retrieval and embedding work, evaluation harnesses that keep the results honest, and backend architecture built to last.

Education

University of California, Berkeley Expected December 2027

Bachelor of Arts - Data Science

MinorComputer Science

Santa Monica College June 2023

Associate of Science - Computer Science

HonorsMagna Cum Laude / High Honors · Phi Theta Kappa

Curriculum

Relevant Coursework

CS 61AStructure & Interpretation of Programs

CS 61BData Structures

CS 61CMachine Structures

CS 70Discrete Math and Probability Theory (Planned)

CS 170Algorithms (Planned)

CS 182Deep Learning (Planned)

CS 186Database Systems

CS 189Introduction to Machine Learning

EECS 183Natural Language Processing (Planned)

DATA 8Foundations of Data Science

DATA 100Principles & Techniques of Data Science

DATA 140Probability for Data Science

MATH 54Linear Algebra & Differential Equations

Experience

Software Engineer (Intern) July 2026 – Present

WithMemberly

Build an LLM recommendation assistant on the Anthropic API with streaming responses, live web search, and producer-context retrieval.
Build and ship user-facing features for WithMemberly's membership platform across a Next.js/React frontend, using TanStack Query for server-state management and a reusable Radix UI component library to keep the interface consistent and accessible.
Develop RESTful API endpoints in a layered Node.js/Express architecture (routes → controllers → services with DTO validation) backed by Prisma and Postgres, including Stripe integration for member and producer payment flows.
Maintain and extend CI/CD and deployment on AWS, using GitHub Actions and CodeDeploy to ship Dockerized services to EC2, with Sentry instrumentation for production error monitoring.

Research

Global Opportunity Lab | URAP September 2026 – Present

Mentor: Prof. Joshua Blumenstock

Applying deep learning to satellite imagery for Sand Mining Watch, developing semi-supervised models to detect illegal sand mining in India under limited ground-truth labels.

NIST Center for Neutron Research August 2026 – Present

Mentor: Dr. William Ratcliff

Building a neural retrieval system that matches noisy X-ray diffraction signals to a database of known crystal structures, training PyTorch dual encoders with contrastive and ranking losses and serving embeddings via ANN vector search.
Benchmarking learned embeddings against classical similarity and optimal-transport baselines, using leakage-aware train/test splits to prevent data leakage across related samples.
Evaluating robustness to distribution shift, noise, missing features, and mixed-source inputs, with open-set rejection for out-of-distribution queries and interpretable match explanations.

A line through my work

Projects

July 2026

GBTutor

An LLM agentic tutor with a from scratch retrieval core and evaluation harness, deployed as a three tier ML backed web app, that learns your course from your own materials and teaches you to recognize and attack exam problems.

AI-AgentPythonRAGPyTorchLLM

View on GitHub →

June 2026

Interactive Portfolio

This site. A from-scratch, single-page portfolio in HTML, CSS, and JavaScript: a scroll-driven full-page section deck with a custom galactic theme and hyperspace transitions, and CSS-animated UI, built with no frameworks.

JavaScriptHTMLCSS

May 2026

Build Your Own World (BYOW)

An Animal Crossing themed 2D island-world engine in Java built from-scratch — seeded procedural generation for reproducible worlds, BFS pathfinding across a 4,000-tile grid, and a full game loop with undo/redo, save/load, music, and a live HUD.

JavaPsuedorandom GenerationBFSFull-Stack

View on GitHub →

April 2026

Housing Price Prediction & Fairness Analysis

An end-to-end scikit-learn regression pipeline over 200K+ Cook County records reaching 0.61 log-RMSE, with from-scratch k-fold CV and a custom MAPE loss that exposed bias standard metrics hid.

Pythonscikit-learnPandasNumPyFeature Engineering

View on GitHub →

April 2026

WordNet Semantic Search Engine

A full-stack semantic search engine traversing an 82K-relationship WordNet graph via BFS, indexed into HashMaps/TreeSets at startup for O(1) lookups and NGram-ranked multi-word queries.

JavaGraphsFull-StackHTMLJavaScript

View on GitHub →

March 2026

Logistic Regression Spam Classifier

A logistic-regression spam classifier in scikit-learn with 60 engineered features that lifted accuracy from 76% to 94%, tuned via 5-fold GridSearchCV and judged on precision, recall, and false-positive rate.

Pythonscikit-learnPandasRegressionModel Training

View on GitHub →

March 2026

Percolation

A union-find percolation simulation on an N×N grid, built test-first and paired with Monte Carlo trials to estimate the percolation threshold.

JavaUnion-FindMonte Carlo

View on GitHub →

November 2025

Climate Change

A Python analysis of historical temperature and precipitation data, building 99% bootstrap confidence intervals to surface warming trends and running permutation tests to assess whether drought years differ in rainfall.

PythonNumPyBootstrap SamplingHypothesis Testing

View on GitHub →

November 2024

ChatsApp

A real-time client–server chat app in JavaFX over TCP sockets, with a dedicated reader thread marshaling updates via Platform.runLater and a styled, auto-scrolling bubble interface.

JavaJavaFXSockets

View on GitHub →

Toolkit

Skills

Languages

C++CJavaPythonSQLJavaScriptHTMLCSSSwiftTypeScript

Tools and Frameworks

ReactNext.jsNode.jsExpress.jsPrismaTanStack QueryRadix UIJavaFXGitGitHubDockerAWS EC2AWS AmplifyAWS CodeDeployAWS IAMSupabaseStripeSentryGoogle ColabHuggingFacePostgreSQLJupyterDatabricks

Libraries

PyTorchPandasNumPyMatplotlibSeabornSciPyScikit-Learnsentence-transformers

Contact

Let's build.

Email GitHub LinkedIn