Working towards frontier AI that’s both capable and interpretable.
Selected work
Three write-upsNASA/JPL · AI & Cybersecurity Intern · 2026
Teaching an agent that severity isn’t difficulty
An agentic LLM framework for vulnerability remediation in the Deep Space Network. It pulls context through MCP, returns schema-validated and cited recommendations, and ranks fixes by what they’d cost the network. A person approves every one.
Read the write-upPersonal project · Self-hosted
I gave my AI a memory of everywhere I’ve been
MCP servers on my home server that let ChatGPT and Claude answer questions from years of my own Google Maps Timeline, plus the undocumented backup format that hid most of it.
Read the write-upOpen source · Fork of ynbh/canvasmcp
Letting my AI read my coursework
I extended an open-source Canvas LMS MCP server with Firefox session auth and inline file reads, then ran it headless on a Linux server so any assistant can see what’s due.
Read the write-upExperience
Full CV-
Jun – Aug 2026
Pasadena, CA
AI & Cybersecurity Intern · NASA Jet Propulsion Laboratory
Designed and built an agentic AI application that analyzes cybersecurity vulnerabilities in NASA’s Deep Space Network and generates remediation recommendations.
How it works -
May – Aug 2025
McLean, VA
Machine Learning R&D Intern · MITRE
Researched, prototyped, and delivered machine learning solutions, weighing hardware, framework, and model tradeoffs, ending in a sponsor-facing deliverable and an end-to-end demo. Built a repeatable process for adding ML to existing projects.
-
Jun 2023 – Jan 2025
Leavenworth, KS
Research & Development Intern · Cornerstone Integration
Designed, prototyped, and demonstrated integrated hardware and software systems, working with vendor engineering teams across AI, computer vision, embedded and IoT, cloud, networking, and security.
Education
B.S. Computer Science · B.S. Mathematics
Recognition
- CyberPatriot National Finalist, twiceTop 0.1% of 5,264 teams nationally
- FIRST Tech ChallengeProgrammed driver control and a vision-guided autonomous mode
Rabbit holes
Things I couldn’t leave alone- 01
Why did my timeline stop on August 4?
Google splits encrypted backups into parts listed in an undocumented field, and nothing fetched past part one. I rebuilt ground truth in an Android emulator to prove it. All 8,551 segments came back.
- 02
Can my AI know what’s due this week?
I forked a Canvas LMS MCP server, taught it to read Firefox sessions and hand files straight into the chat, and run it on my home server.
- 03
Where do you test location code without leaking your location?
In the middle of the Sahara. Every test in my location stack uses made-up coordinates there.
What’s next
InterpretabilityFrom the outside in, to the inside out.
At JPL I made an LLM dependable from the outside, with schemas, validation, citations, and a person signing off. Now I want to work on the inside: mechanistic interpretability, representation learning, and model monitoring. That means understanding what capable systems are actually doing, so they can be made reliable and controllable.
I’m looking for research and research-engineering roles, internships, and collaborations.
Let’s talk.
Email is the fastest way to reach me, for roles, research, or just to talk shop.
Outside of work, I’m usually traveling. More about me