A whole book. A coherent process.
An AI-assisted educational-book authoring system that connects research, an editable outline, text and image generation, and human review.
Built at Innovate Nepal Group. This case study shares my contribution and the product architecture at a high level.
- Audience & learning goals
- Chapter progression
- Section structure
Give every chapter a purpose.
Research and an editable outline establish the book’s direction before content generation.
An illustration of the authoring flow, using sample page layouts.
A book needs more than good paragraphs.
An educational book has to work across chapters: the material needs a purpose, a sensible progression, an appropriate level, and supporting visuals. At Innovate Nepal Group, I architected a multimodal book-generation system that brings those concerns into a guided authoring process.
Plan the structure first.
The workflow starts with the book’s brief and research, then develops an editable structural outline before generating the full content. That skeleton gives chapters a shared frame of reference and lets an author review the direction while changes are still inexpensive. Audience and learning goals inform the plan instead of being added after a draft already exists.
Coordinate the words and the images.
Text generation and visual generation serve the same book. The system combines LLM-driven writing with image-generation capabilities inside a structured content workflow. A shared plan supports continuity across sections and helps keep illustrations connected to the material they explain.
Build an authoring backend around the models.
My work connected the generation process to persistent project state and background execution. Book generation is a long-running operation, so the backend has to support progress, saved work, and targeted revisions. Authors can regenerate a chapter or a smaller part of the content without treating every change as a new book.
- Separate interactive API requests from longer generation work.
- Persist the outline and generated content throughout the authoring process.
- Support review and selective regeneration at useful editorial boundaries.
- Keep text and visual assets connected to the book they belong to.
Keep the author in the process.
The product is an assisted authoring workflow. Research, structure, generated material, and final review remain visible stages that people can inspect and revise. That matters especially for educational content, where a fluent draft still needs a subject expert’s judgment about accuracy, suitability, and coverage.
The engineering contribution.
This work brought together multimodal AI, workflow orchestration, and backend product design. The central architectural decision was to establish structure before generation, then make the resulting material persistent and editable. That turned model outputs into an authoring process people could work with over time.
Let’s build
something that works.
Have an AI problem, a backend challenge, or a team I should meet? I’d like to hear about it.
Based in Kathmandu. Open to a good conversation.