Hi, I'm Victoria. I teach machine learning at San Francisco State University, where I designed the intermediate ML course for the PINC program, and I build open-source tools — mostly for reading, note-taking, and personal infrastructure. Based in San Francisco.
Follow me on LinkedIn, or see some of my work on GitHub!
In my free time, I like reading, playing Paradox map games, going on walks around the city, and playing with my terminal.
ResumeMachine learning practitioner and teacher: I designed and lecture CSC 411 (Intermediate Machine Learning) at San Francisco State University, after six months in R&D data operations for a physical-AI company. B.S. Computer Science, SFSU; U.S. Air Force veteran.
Victoria Barnett
[email protected] · 415-930-2560 · San Francisco, CA
vb4r.com · github.com/victoria-riley-barnett · linkedin.com/in/victoriabarnett0
Software Engineer · Machine Learning Engineer · Systems
Summary
I build practical software and machine-learning systems, and I teach others how to understand and improve them. At San Francisco State University, I designed and currently teach a 15-week intermediate machine learning course that takes students from matrix multiplication and gradient descent to a working transformer encoder. Before and alongside teaching, I worked in R&D operations for a physical-AI company, coordinating operators, devices, telemetry, and dataset quality into alignment with the ML team. I enjoy tracing problems to their source, making complex systems easier to use, and leaving behind tools and documentation that others can rely on.
Experience
Lecturer and Course Developer — CSC 411: Intermediate Machine Learning
San Francisco State University, San Francisco, CA · May 2026 – Present
- Designed and teach the first offering of a fifteen-week course for the PINC interdisciplinary computing minor, with no calculus prerequisite.
- Teach students to derive and implement gradient descent, backpropagation, multilayer perceptrons, and attention from scratch before using PyTorch and Lightning to build a working transformer encoder.
- Created 18 lecture and lab notebooks, rubrics, and a checkpointed capstone; built notebook-to-Markdown conversion and PDF-export tooling to support course delivery.
- Replaced a 12 year old curriculum with a FastAi / SmolML-inspired curriculum that emphasizes architectural understanding, evaluation, debugging models, and extensive hands-on toy model and Pytorch work, rather than using pre-built architectures. Supplements with historical and contemporary research papers, including Domingos’ “12 Lessons,” and Vaswani et al.’s “Attention is All You Need.”
R&D Operations Coordinator (Contract, via Lumicity)
Confidential AI/robotics client, San Francisco, CA · February 2026 – August 2026
- Supervised the data operation behind a physical-AI training pipeline, coordinating hundreds of collection devices, the operators running them, and the QA standards used to accept or reject data.
- Used SQL and Retool to trace 80% of logged device faults—approximately 40,000 events—to a single upstream cause, and helped analyze and fix the underlying issue in software to operator interaction.
- When dataset diversity collapsed mid-season, pinpointed the collection process as the cause, and experimented with new collection strategies to restore it.
- Turned operator and telemetry findings into collection, system, hardware, and process improvements across the startup, including a rapid quality-assurance pipeline for external partner data.
Instructional Assistant — CSC 511: Protein Modeling with Deep Learning
PINC Program, San Francisco State University · December 2025 – May 2026
- Led twice-weekly recitations for approximately 40 life-science students building protein-property-prediction models in PyTorch and Lightning.
- Wrote Jupyter notebook tutorials for AlphaFold2-inspired architectures and helped students debug model logic, CUDA, GPU-memory, and environment failures.
- Designed and graded assignments, quizzes, and exams, and provided feedback on student code and reports.
Chapter Co-Chair, Technical Infrastructure (Volunteer)
Democratic Socialists of America, San Francisco · 2021 – 2024
- Ran technical infrastructure without an engineering team: three servers across DigitalOcean and AWS, fifteen Docker services managed with Portainer, Auth0 single sign-on, and CRM and publishing integrations.
- Trained non-technical volunteers on data pipelines and campaign tools including NationBuilder, Action Network, Mobilize, CallHub, and Google Workspace.
Crew Chief, then Inspector — Munitions Systems
United States Air Force · 2016 – 2021
- Led maintenance and production teams of 2–8 while maintaining and inspecting aircraft munitions systems.
- Found data-integrity errors across databases tracking $400 million in guided munitions, traced the errors to their source, and drove a technical-data correction that propagated globally in coordination with Boeing and Raytheon engineers.
- Wrote SQL report scripts over an Access reporting layer, replacing a weekly report that took three to four hours to compile by hand.
- Authored, managed, and maintained more than 120 technical and operational documents to formal standards.
Selected Projects
Tailscale for KOReader — VPN Support for ARM E-Ink Readers
- Ported Tailscale to ARM e-ink readers (Kindle, Kobo, PocketBook, reMarkable) and wrote the Lua plugin that integrates it into KOReader, which previously had no VPN support.
- Maintain the project, which has shipped seven releases, merged 11 pull requests from outside contributors, and drawn 223 GitHub stars and 24 forks.
- Root-caused a routing failure to forced userspace networking and moved the project to kernel TUN with a userspace fallback; also fixed an installer that could report success after failed extraction.
Cairn — Desktop IDE and Orchestration Tool
- Building a desktop IDE around a small programming language, a graph-backed file system, and an experimental partial-evaluation engine.
- Modeling org-mode documents and live program state as first-class, queryable data.
- Developing dispatch, evaluation, and context-management tooling for combined human-ML workflows within the IDE.
SerialPM — Visual project-management platform
- Led a five-person capstone team building Node/Express and React/TypeScript microservices with MySQL, Redis, Docker, AWS, and CI/CD.
- Authored 16 merged pull requests spanning schema, API, and dashboard work.
Alzheimer’s Screening Classifier — Recall-first machine-learning project
- Led a three-person team building a Random Forest classifier in R, using cross-validated grid search to optimize sensitivity rather than accuracy.
- Shifted the voting cutoff toward detection, reducing missed diagnoses by 68% at a cost of 0.65 points of accuracy.
Auto Child Education — Europa Universalis V mod
- Created one of the ten most-subscribed Europa Universalis V mods on the Steam Workshop, with about 25,000 subscribers.
- Kept the mod compatible through three game updates and shipped fixes for trait detection and performance.
Open Source Contributions
- D2R Reimagined: Contributed nine merged pull requests to a 150-star community game mod, covering item and recipe data, balance tuning, localization, and diagnosed bug reports.
- DSA Planet: Authored the “DSA Light” theme and landed six merged pull requests in the national organization’s RSS aggregator.
- md-advanced-tables: Fixed a formula reference in a 190-star Markdown table editor (merged upstream).
Education
B.S., Computer Science — San Francisco State University, December 2025
Data structures and algorithms, operating systems, compilers, computer architecture, database systems, software engineering, system administration, and AI ethics. Prior coursework at City College of San Francisco; Dean’s Honor List, Fall 2022.
Skills
Machine Learning and Data: PyTorch, Lightning, scikit-learn, pandas, NumPy, R, neural architectures from scratch, LLM evaluation, failure taxonomies, agent transcript scoring, model evaluation, dataset QA, acceptance standards, telemetry pipelines, embeddings, graph inference, retrieval, knowledge graphs, SPARQL
Programming Languages: Python, TypeScript, JavaScript, C, C++, Rust, Java, Common Lisp, SQL, R, Lua, Fennel, Bash
Web and Applications: React, SolidJS, Node/Express, REST APIs, Astro, PostgreSQL, PostGIS, MySQL, SQLite, Redis, MapLibre, GeoJSON
Systems and Infrastructure: Linux, systemd, Docker, Docker Compose, Portainer, AWS, DigitalOcean, Cloudflare, CI/CD, release automation, cross-compilation, embedded ARM, WireGuard/Tailscale VPN
Tools: Jupyter, Retool, Auth0, Google Workspace, LaTeX, Cargo, uv, Make, SBCL/ASDF
TeachingLecturer for CSC 411 (Intermediate Machine Learning) in the PINC program at SFSU: teaching the first offering of the course I designed, from matrix multiplication to a from-scratch transformer encoder.
I teach machine learning at San Francisco State University, in the PINC program: a computing minor built for biology, biochemistry, and chemistry students. I designed its intermediate machine learning course and now teach its first offering.
CSC 411: Intermediate Machine Learning for Interdisciplinary Data Scientists
The course takes students with no calculus, linear algebra, or machine learning background and walks them, principle by principle, from matrix multiplication to a working transformer encoder built from scratch. Each operation of deep learning is derived and implemented by hand before the libraries arrive: a training loop, an MLP layer, an attention head, composed into the architecture behind modern language models.
Format: twice a week, a lecture and a hands-on lab. The weekly notebook is the unit of work, with only a small graded check at the end of each one. Students run a portfolio project on a dataset of their own choosing, with checkpoints across the term and a written self-analysis at the end. They also read and discuss real machine learning papers, matched to the level of model they have just built.
Principles that shape the course:
- Meet students where they are. No prerequisites assumed; the math is taught as it is needed.
- Derive before you delegate. You can only trust a model you can take apart.
- Grade for understanding, not mechanical completeness. Clarity about what is scored is itself a form of care.
- The artifact is the point. Students leave with a model they built and a project that is theirs.
Before teaching this course I was an instructional assistant for CSC 511 (Protein Modeling with Deep Learning), leading recitations for life-science students building protein models in PyTorch and Lightning.
Work HistoryProfessional experience spanning university teaching, R&D data operations for an AI/robotics company, network contracting, volunteer technical leadership, and U.S. Air Force munitions systems.
| Company | Position | Key responsibilities | Timeframe |
|---|---|---|---|
| San Francisco State University | Lecturer, CSC 411: Intermediate Machine Learning (PINC Program) | Teach the first offering of a course I designed: fifteen weeks from matrix multiplication to a from-scratch transformer encoder, with weekly lectures and hands-on labs, and a checkpointed portfolio project on each student’s own dataset. | Fall 2026 — present |
| Confidential AI/robotics client, San Francisco (contract via Lumicity) | R&D Operations Coordinator | Supervised the operators and data-collection devices producing model training data; managed scheduling, training, device readiness, and dataset QA. Redesigned collection around diversity after a key dataset metric collapsed. | Feb 2026 — Aug 2026 |
| San Francisco State University, PINC Program | Instructional Assistant, CSC 511: Protein Modeling with Deep Learning | Led twice-weekly recitations for ~40 life-science students building protein models in PyTorch and Lightning; wrote the course’s Jupyter notebook tutorials; debugged alongside students from model logic down to GPU-memory and environment issues. | Dec 2025 — May 2026 |
| Property management company, San Francisco | IT Contractor (hourly) | On-call network support dispatched per property: installed and configured mesh Wi-Fi, resolved ISP hookups (Sonic, Xfinity). | Dec 2023 — Feb 2025 |
| Democratic Socialists of America, San Francisco | Chapter Co-Chair (volunteer) | Ran the chapter’s technical infrastructure end to end (servers, Docker microservices, Auth0, CRM and publishing integrations); trained volunteers on data-pipeline and campaign tools. | 2021 — 2024 |
| U.S. Air Force | Munitions Systems (Crew Chief, then Inspector) | Led 2–8 person maintenance and production teams; as inspector, discovered hazard-tracking errors across databases covering $400M in guided munitions, triggering a global technical-data update; authored and maintained 120+ technical-operational documents; built SQL and Access reporting tools. | 2016 — 2021 |
Notes on Roles & Transferable Skills
- Teaching & Curriculum Design: designed and built a university machine learning course from scratch, and taught it to interdisciplinary students with no math prerequisites; led recitations and wrote course notebooks for a deep learning course for life-science students.
- Leadership & Teamwork: supervised small technical teams, coordinated cross-organizational workstreams, trained personnel and volunteers, and led student project teams.
- Documentation & Communication: authored formal technical documents and operational reports; writes about machine learning, systems, and teaching.
- Systems Thinking: experience maintaining safety-critical tracking systems, debugging multi-system workflows, and automating repetitive tasks; comfortable at every level from GPU memory to database reporting layers.
- Technical Growth: ongoing work in machine learning, curriculum development, and open-source tooling.