I'm Shivnath, a software engineer at ISG eSolutions where I mostly work on AI infrastructure. RAG pipelines, PDF extraction with LLMs, private AI deployments. The kind of stuff that sounds straightforward until you're three days into a context window bug at 1am. On the side I do independent AI research. Two arXiv papers so far, one on continual learning (LACE) and one on 4-bit quantization. I also built DiskLLM, a library that cuts LLM RAM usage by ~57x at long contexts. It's on PyPI if you want to try it. I like first-principles thinking more than I like tutorials. If you want to actually understand how something works rather than just run the starter code, that's probably where I'm most useful.