Hi, I am Sankalp. I go by dejavucoder on X. I am into AI engineering and applied LLM research. I maintain a pseudonymous profile on X but you can find mo...
Table of Contents Intro Contest in short Problem intro Why this problem is auto-research-able Learning Enough to Ask Better Questions (Optional) Math for QR decomposition: Householder reflections Make serial work small with the help of the blocked Householder algorithm Other challenges Codex-maxxing Kernel progress breakthroughs Breakthrough ideas Introducing idea diversity to escape the local…
Lore It was mid-May. I was a couple of weeks into catching up on post-training literature and learning reinforcement learning for LLMs in a more bottom-up way. Having done some basic experiments, I was unable to come up with an idea that had the allure to possess me for a week or two. I waited as if I was entitled to receive the idea from the muses. I wondered - if I can be creative in other areas…
There was a lot of discourse around Peter's recent tweet. I think the main driver behind this was most people not taking the time to think about Peter's context before comparing it with their own. In this post, I will talk about this "context". Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents. — Peter…
Fable-5 High You can find Opus 4.8 output after this. From SGD to Muon: An Incremental Tutorial This tutorial builds up to the Muon optimizer one idea at a time. Each section adds exactly one concept on top of the previous one, so by the time we reach Muon, every design decision should feel inevitable rather than mysterious. The path we'll take: Gradient descent, and weights as matrices Momentum…
Kpop I started listening more to kpop over the last year. Most of the kpop songs I listen to lean towards being high energy and full of electronic and synth sounds. I think they work as a nice stimulation along with the visuals, choreography and creative concepts in their videos. If I am feeling low or dull, they often help me cross a certain activation energy, after which I become more…
A deep dive into Claude Code 2.0 features, Opus 4.5 workflows, and context engineering. Learn sub-agents, MCP servers, hooks, skills, and practical tips to boost your AI-assisted coding productivity.
A deep dive into prompt caching - practical tips to improve cache hits and how vLLM's paged attention enables KV-cache reuse across requests via automatic prefix-caching
UPDATE: If you want to understand how prompt caching works under the hood, read my blog how prompt caching works - paged attention and prefix caching plus practical tips . This blog is an extract out of that just covering the tips part for readability reasons. Prompt caching is when LLM providers reuse previously computed key-value tensors for identical prompt prefixes, skipping redundant…