Reflections and deep dives on computer architecture, LLM security, and systems engineering.
Reflections on ranking in the top 1.5% nationwide among 40,700+ final-year CS/IT students across 119 universities. Why industry standardized testing reveals a widening gap in first-principles understanding — from cache hierarchies and bitwise manipulation to algorithmic invariants.
Benchmarking throughput, concurrency, and time-to-first-token (TTFT) when serving dense open-weight models on modern unified CDNA3 architecture using vLLM and PagedAttention kernels.
Key findings from the Certified LLM Security Professional designation: How indirect prompt injections break traditional sanitize-and-filter methods, and why zero-trust tool authorization is mandatory for autonomous agents.
Why modern software needs a "fade-out" design pattern that empowers users to outgrow apps rather than trapping them in gamified engagement loops. Lessons from building privacy-first spatial recall and attention drills.