Fiction
Director plans the ending first, then parallel scene writers and chapter reviewers produce coherent long-form novels (~300 pages).
Local LLM · Multi-agent · Book-length output
Seven publishing modes—fiction, textbooks, hardbacks, language courses, and media-rich STEM PDFs—powered by hierarchical local LLM agents.
0
Book publishing modes
0
Words per novel draft
0
Wall-clock on M2 Ultra
0
Completion with retry logic
Specialized agents share a single canon—synopsis, curriculum, or language guide—then draft in parallel with tier-specific concurrency tuned for local Ollama pools.
Fiction · 4 tiers
Textbook, hardback & language · 6 tiers
Media-rich textbooks · 7 tiers
One orchestration engine—from novels and STEM textbooks to university hardbacks, bilingual courses, and media-rich PDFs with Ollama figures, embedded video, and ManimRedux animations.
Director plans the ending first, then parallel scene writers and chapter reviewers produce coherent long-form novels (~300 pages).
Instructional books with curriculum outlines, section-level lessons, end-of-chapter problems, and a separate answer key.
Full-length university texts with rigorous curriculum, multi-step worked examples, print-ready 7×10″ hardback PDF layout.
Vocabulary tables, example sentences, typed drills (fill-in-the-blank, multiple choice, matching), and model answers.
Illustrated STEM texts: Ollama x/z-image-turbo generates PNG figures sprayed at pseudo-random paragraph positions in each section.
Digital PDFs with embedded MP4 demonstrations via \includemedia. ffmpeg placeholders today; text-to-video hook reserved.
Animated diagram MP4s rendered by ManimRedux (Rust Manim port), transcoded with ffmpeg, embedded in LaTeX for Adobe Reader.
Read the full industrial-track paper with algorithms, benchmarks, and engineering findings—without digging through implementation source.