I have spent the last six years building machine-learning systems, across applied NLP and LLMs, time-series, anomaly detection, and graph models. I take them through the full lifecycle, from framing a problem to a model that runs reliably in production, with a focus on evaluation, on whether a change earned its place.
Recently: an SDXL LoRA fine-tuned on WPA-era posters, with an eval harness (CLIP, FID, KID) and an MCP server: wpa-poster-sdxl
Also published: Graph Neural Network based Root Cause Analysis (IEEE NOMS 2022).