Kevin Marin
Seeking entry-level Data Science / Software Engineering roles
B.A. Data Science, Claremont McKenna '26 · Open to relocation · Remote (US)
A reliability and evaluation harness for enterprise AI agents, benchmarked on ServiceNow's WorkArena: +30 points task success and 61% fewer output tokens on held-out tasks, with every number traceable to a logged episode.
Three models I trained in PyTorch (MNIST digits, 75-class Quick, Draw! doodles, and a character-level GPT), exported to ONNX and running live in the browser: play Pictionary against the doodle model, x-ray the digit CNN layer by layer, or watch the GPT's attention heads think.
An AI health CLI: a conversational LLM intake feeds a logistic mortality-risk model trained on public NHANES data, then writes a lifestyle report and daily schedule. Privacy-first, runs fully local.