Now

Building

• hermes-curriculum - A knowledge-graph curriculum engine that decides what an AI tutor should teach next: FSRS spaced repetition, prerequisite gating, exam questions grounded in real course material. It grew out of Book Summarizer once I realized summaries weren't the point - knowing what to study next was.
• A personal agent that actually does things - I run an always-on assistant on my own server: it drills me daily for exams over Telegram, curates systems-and-AI papers into the news section of this site, and keeps growing new jobs.

Learning

• Concurrent and Parallel Programming - Currently preparing for the next exam and sharpening my mental models for concurrency, synchronization, and parallel execution.
• GPU programming - Working through Programming Massively Parallel Processors, with a 3,000-question curriculum my own engine generated from the book. Eating my own dog food.

Questions I keep returning to

• Human reasoning + machine retrieval - LLMs can retrieve and reorganize an absurd amount of information, but that is not the same as reasoning. What should the system between those two abilities look like? I am not claiming to know; Hermes is one way I am trying to find out.
• Constraints before generation - I do not trust a model with an end-to-end project and no standard outside itself. In a recent legal-editor replacement, the useful check is how many items match the old system byte for byte. The model gets room to work, but not room to declare itself correct.

Recently

• Cybersecurity: 29/30 - Completed the exam with a curriculum generated by hermes-curriculum. A useful result for the degree, and an even better validation that the system works on real material under real pressure.

Reading

• Rust Atomics and Locks (Mara Bos) - because I don't really trust my mental model of memory ordering yet.
• AI Engineering (Chip Huyen) - the missing manual for the LLM-systems work I keep gravitating toward.

Open to

I am always looking for the most interesting challenges I can find, preferably with interesting technology behind them. I am open to work that lines up with those interests, stretches what I know, and leaves room to learn.

Life

Finishing the master's while working full-time, which is exactly as much fun as it sounds. Thinking hard about what comes after the degree - the interesting problems I want to work on live at the intersection of distributed systems and AI infrastructure, and I intend to go where they are.