Journal
Short essays on building interfaces, shaping motion, and putting machine learning to work without the hype.
A practical framework for turning vague AI requests into clear, testable instructions—with reusable templates, examples, and a quality checklist.
A realistic workflow for adapting gpt-oss-120b with LoRA—from deciding whether to fine-tune through data design, memory planning, evaluation, and deployment.
A production-minded guide to extracting structured data from complex documents with traditional OCR, multimodal LLMs, strict schemas, validation, and human review.
A practical blueprint for dependable AI pipelines: explicit contracts, reproducible training, safe deployment, useful monitoring, and simple operations.