Selected work
I build agentic developer tools, learning products, and low-level systems. These projects best show how I work: start with the human problem, make the logic explicit, and verify what the system actually does.
Agentic + adaptive
Systems shaped by a codebase, a learner’s behavior, or the assumptions behind an interface.
A Claude Code orchestration plugin that profiles a repository, selects the lightest reliable workflow, and builds validated task graphs for larger work. Independent review and verification keep agentic complexity accountable.
A browser editor that marks Spanish spelling and grammar issues in place, explains corrections, and supports follow-up questions with an AI tutor. DOCX import and export keep the help inside the writing workflow.
A focused Spanish practice tool that tracks accuracy by concept, tense, and verb family, then prioritizes weak areas without repeating the same exercise endlessly.
A final-project exploration of AI, self-reflection, and the value of a well-timed question.
Technical foundations
Coursework where correctness lives below the interface: circuits, memory, probability, and graph structure.
A functional Logisim processor with a custom RISC-style instruction set, ALU, registers, control unit, program counter, and memory interface. Nineteen ROM programs verify its behavior.
A complete brick-breaker game written in MIPS assembly, with paddle input, collision detection, bitmap graphics, multiple game states, and custom NES-style text.
A Java graph algorithm that minimizes an attacker’s probability of detection. It transforms multiplicative probabilities into edge weights, then applies Kruskal’s algorithm and optimized union-find.
More context