I build LLM applications and agents that have to survive real users — retrieval that stays honest about what it does not know, tools that fail loudly instead of quietly, and the unglamorous plumbing around both. Six years in, mostly on the application side. Lately I have been going a layer down: inference serving, and the operational work of keeping models running.

• Top K Frequent Elements: Why Eviction Is Safe _2026-08-20
• Group Anagrams: Choosing a Canonical Key _2026-08-18

_{More at hyeseonko.github.io}

602 problems solved

_{Solutions live in hyeseonko/LeetCode.}

• hyeseonko/hyeseonko.github.io — pushed 1 commit _2026-08-20
• hyeseonko/leetcode-archive — pushed 1 commit _2026-08-18
• hyeseonko/hyeseonko — pushed 1 commit _2026-08-17
• hyeseonko/LeetCode — pushed 1 commit _2026-08-17
• hyeseonko/PaperSurvey — pushed 1 commit _2026-08-17

_{The cards and lists above are rebuilt daily by a GitHub Action. scripts/build.py draws every SVG itself and talks only to the GitHub API and my own feed, so nothing here breaks when someone else's card service is rate limited.}