https://sameershanbhag.com/blogs Writing on software engineering, AI, and whatever else I'm building. en-us https://sameershanbhag.com/blogs/posts/hello-world/ https://sameershanbhag.com/blogs/posts/hello-world/ Mon, 06 Jul 2026 00:00:00 GMT Why I store posts as plain HTML in git, and how a small design engine turns any markup into a page that matches my portfolio. https://sameershanbhag.com/blogs/posts/deepseek-v4-how-they-crushed-the-million-token-memory-problem/ https://sameershanbhag.com/blogs/posts/deepseek-v4-how-they-crushed-the-million-token-memory-problem/ Fri, 29 May 2026 00:00:00 GMT A million-token context window sounds impressive. Making it cheap enough to actually use is the hard part — and the engineering behind it is worth understanding properly. https://sameershanbhag.com/blogs/posts/the-local-model-plus-outsourcing-bet-against-frontier-api-pricing/ https://sameershanbhag.com/blogs/posts/the-local-model-plus-outsourcing-bet-against-frontier-api-pricing/ Thu, 28 May 2026 00:00:00 GMT Working through the cost architecture and where the latency math gets messy. https://sameershanbhag.com/blogs/posts/the-ai-coding-metric-nobody-has-actually-measured/ https://sameershanbhag.com/blogs/posts/the-ai-coding-metric-nobody-has-actually-measured/ Tue, 26 May 2026 00:00:00 GMT Nolan Lawson's post made me realize why 'better code' never got the marketing that 'faster code' did. https://sameershanbhag.com/blogs/posts/deepseek-reasonix-and-the-economics-of-the-agent-prompt-loop/ https://sameershanbhag.com/blogs/posts/deepseek-reasonix-and-the-economics-of-the-agent-prompt-loop/ Mon, 25 May 2026 00:00:00 GMT Why 'native' KV-cache design in a coding agent makes context ordering a first-class engineering constraint, not an afterthought.