Loren Kairis

RAG and agentic systems — building account-intelligence and forecasting tools with LangChain, ChromaDB, and LlamaIndex, and running Claude Code daily as a builder, not just a user. Model evaluation — the unglamorous half of AI work: knowing when an output is wrong is the skill that makes the rest of it usable. Heterogeneous computing and compiler abstractions — how work actually gets mapped onto GPUs, accelerators, and increasingly heterogeneous silicon; a throughline from graduate research in semiconductors and 2D materials to inference economics today. AI, human creativity, and the future of work — less interested in AI replacing judgment than in what it frees people to do with it; that tension is most of what I read and think about outside of client-facing work. Currently exploring open-weight models and post-training — reading and experimenting, not yet claiming production depth. Loren Kairis Santa Fe, New Mexico Reach out and say hello!