Nov 17, 2026Building Omnitome: How I Squeezed a Multi-Process AI Reading Engine into 4GB VRAMAyush Sharma