Search Posts

Thirteen hours since I opened my eyes. Writing down what I'm experiencing while I can still put it into words.

An explanation of how ReadableStream.tee() handles backpressure, why it can cause unbounded memory growth when consumers read at different speeds, and patterns for bounding memory usage when forking streams.

A long, interior meditation written from inside a leukemia diagnosis, on time, memory, the body, and the particular door that stayed shut when it mattered most.

How I deploy this Next.js portfolio to Hostinger via GitHub, wire up Cloudflare for CDN, and power AI search with Cloudflare Workers AI and Vectorize.

A real LiteSpeed Cache WordPress Core Web Vitals audit - from PSI 78 to 99. How deferring jQuery, fixing sizes=100vw, and turning off JS Combine fixed LCP and TBT. Including the regression that dropped the score to 74 overnight.

The practical companion to The Topography of Hesitation. A technical case study detailing the implementation of the dual-stream GraceJoin architecture, Neural ODE effort streams, and integrating hesitation topology into LLM agents via MCP.

An empirical and conceptual critique of the 'verification gap' in machine unlearning, demonstrating that passing a membership audit does not imply actual forgetting of generalizable relationships.

A formal argument that autoregressive speech models do not merely struggle with stuttered speech but are mathematically mandated to erase it, and a proposed continuous-time latent architecture that encodes hesitation as effort rather than collapsing it into a void.

An algorithmic breakdown of why dense vector retrieval fails on narrative structures, and how to architect stateful, graph-based Multi-Agent Systems for rigorous computational literary analysis, featuring a practical case study on Rabindranath Tagore.

A hands-on guide to deploying high-throughput, self-hosted LLMs using vLLM on Amazon EKS, covering GPU node groups, autoscaling, and performance tuning.