Hi, I'm Tyler
I work on machine learning and AI safety, from research experiments to production systems. I recently completed projects with Redwood Research on attack selection for AI control and with SPAR on scalable oversight. I also have more than seven years of production software engineering experience, most recently as a senior engineer and team lead. I completed Georgia Tech's MSCS in May 2026, with a focus on machine learning and computing systems.
My path
My undergraduate degree is in Biomedical Engineering. I ended up in software, where I worked on consumer products, enterprise search, data pipelines, developer tooling, and distributed services. That work taught me how to turn prototypes into systems people can rely on and how to make tradeoffs when correctness, performance, and delivery all matter.
Seeing the early results from AlphaFold in 2018 pushed me to study machine learning more seriously. As language models became more capable and agentic, I became interested in a narrower set of questions: what safety evaluations actually measure, how models behave under pressure, and how to build experiments that make those failures visible. I returned to school to add deeper ML and systems foundations to my production engineering background.
Recent research
- AI control with Redwood Research: I studied how an attacking agent's decision to start or abandon an attack changes the apparent safety of control protocols. Selective start/stop policies reduced measured safety by roughly 20–28 points at a 1% audit budget across BashArena and LinuxArena. The project began through MARS 4.0 and continued as a collaboration with Redwood.
- Agent evaluation and scalable oversight with SPAR: In Fall 2025, I built evaluation infrastructure for measuring goal drift in long-horizon coding agents. In Spring 2026, I worked on human-AI oversight and built the FastAPI backend for a participant study run through Prolific.
I also completed ARENA 7.0 in London. For the capstone, I used activation patching, steering vectors, and path patching to study how moral fine-tuning changes attention pathways in Gemma-2-2b.
Selected publications
- Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safety, co-first author. Accepted to the ICML 2026 AI4GOOD and AI-WILD workshops and as a FAGEN poster; under review at NeurIPS 2026.
- When Offline Selectors Cannot Beat the Best Single Model, first author. Accepted to the ICML 2026 DEMO Workshop.
- Three co-authored papers from SPAR, including work on inherited and asymmetric goal drift in coding agents and inference effects in medical allocation.
My full publication list is on Google Scholar.
Industry experience
Knowable
Senior Software Engineer & Team Lead | Jun 2019 – Jun 2024
Directed the technical roadmap for a five-engineer team building a legal knowledge platform. I took RAG-style LLM and OpenSearch systems from data-science prototypes to production over several million contracts, built event-driven services on AWS, and cut the average CI pipeline from roughly 55 minutes to 15. I also maintained a React component library used across three product teams.
Rosetta Stone
Software Engineer | Aug 2017 – Jun 2019
Rebuilt a React and Node marketing platform with page loads about 2.25 times faster. I also built an internal A/B testing system used by roughly 20 non-engineering stakeholders to run hundreds of experiments and maintained backend services for a platform serving millions of learners.
Education
Georgia Institute of Technology
MS Computer Science | May 2026
Focused on machine learning and computing systems. Coursework included deep learning, reinforcement learning, GPU hardware and software, high-performance computer architecture, Advanced Operating Systems, and Graduate Algorithms. Projects included FlashAttention and KV-cache CUDA kernels, GPU simulators, static branch-divergence analysis, and reproducible ML evaluation pipelines.
Virginia Commonwealth University
BS Biomedical Engineering | Dec 2023
Studied engineering, physiology, and the design of systems for medical and biological problems.
What I'm interested in
🧩 Mechanistic Interpretability
Sparse autoencoders, circuits, and tools to open up black boxes so we can understand and steer models.
🎯 AI Safety & Alignment
Building systems that are steerable, reliable, and beneficial by design.
🛡️ Robustness & Evaluation
Empirical evaluation under distribution shift; honesty/helpfulness metrics; reproducibility and measurement.
🔧 Systems for ML
Operating systems, distributed systems, and tooling that make ML research reliable, scalable, and repeatable.
⚡ High-Performance Computing
GPU programming, parallelism, computer architecture, and the memory systems underneath modern models.
📚 Learning in Public
Sharing the journey—methods, failures, and breakthroughs—so others can learn alongside me.
⚖️ Economics
Understanding the economic forces that shape our world.
🧬 Computational Biology
Understanding the biological mechanisms that underlie disease and development.
About This Digital Garden
This is where I keep technical write-ups, research notes, course retrospectives, and project logs. Some pieces are finished articles; others are notes that I update as my understanding changes.
I try to include the method, mistakes, and intermediate reasoning instead of only the final result. If you notice an error or want to discuss something, send me a note.
Built with Astro. Check out the source code.
Sites that inspired this one
- Distill.pub - Interactive machine learning research papers
- Maggie Appleton - The digital garden philosophy and beautiful visual thinking
- Arlen McCluskey - Clean technical writing and site structure
- Eugene Yan - Thoughtful ML content and career insights
- Swyx - Learning in public and developer advocacy
- Yxlow - Minimalist design and thoughtful curation