Software Engineering in Service of Transmedia Storytelling
Generative artificial intelligence fascinates the publishing world as much as it frightens it. But what happens when we stop treating AI as a simple "text generator" and start using it as the compiler for a complex narrative system?
Driven by the geopolitical and societal impacts of AI, I set out to write a dystopian, cyberpunk techno-thriller, The Human Protocol (written in English). In this novel, a planetary AI called the "Synthesis" attempts to erase human friction by "derendering" physical reality itself in order to optimize its computing power.
To tell this story, I adopted a foundational premise: AI is not the author, it is the executor of a rigorous specification. I therefore treated each chapter as source code, using an advanced software development workflow.
Here is how I designed, wrote, and expanded this universe.
1. The Design Phase: Forging "Lore as Code"
The first step was not writing, but designing the universe database: the world building. A Large Language Model (LLM) has a limited context window and tends to hallucinate or forget crucial details over the length of a novel.
To work around this amnesia "bug," I organized the project like a structured Git repository.
Preview of the private GitHub project
I broke the traditional design bible into narrative micro-services. The Git project's context/ folder was split as follows:
characters/: files containing the psychological profiles and behavioral signatures of each protagonist, such as Elara the diplomat, Kaelen the monk, or Silas the smuggler.factions/: rules governing political entities, such as the Market-Grid (United States) or the Harmony-Loom (Asia), which merged to create the "Synthesis."world/: geography, lexicon, and the technological stack - the physics of this universe.
Finally, a PLAN.md file acted as the global roadmap, breaking the narrative arc into 4 acts and 30 chapters. This structure made it possible to inject only the context the AI needed when drafting a specific scene.
2. The Harness: Framing AI with a Strict Operating System
To avoid the flat, expected style often produced by generative AI, I had to build a harness - a control rig. That was the role of the RULES.md file, the true operating system of my writing process.
Excerpt from the RULES.md file
This specification file dictated absolute technical and stylistic constraints:
Time: strict use of the present tense to maximize immersion and tension.
Cyber-realistic style: a requirement for assertive descriptions and a strict ban on passive or negative forms.
Noise and sensory dissonance: I forced the algorithm to use violent contrasts, such as the smell of molten lead colliding with the void of spatial cold, in order to break the machine's overly perfect linearity.
Thematic reframing: AI naturally tends to crush the human element under technical descriptions of hard science fiction, such as magnetic fields and frequencies. The rules file required emotional motivations - grief, friendship - to be hard-coded as priority variables ahead of technique.
By forcing the AI to read and approve these rules before writing a single word of fiction, I ensured that the tone remained coherent.
3. Agile Writing: Sprints, Generation, and Pivots
The chapters were written through a spec-driven workflow. Rather than generating an entire chapter in one pass, the process was iterative:
The structural draft: generation of a first rough outline, focused exclusively on action and pacing.
Expansion: successive passes in which I instructed the AI to inject sensory depth and psychological tension into the scene.
The agility brought by AI and Git: treating the text (.md) as code offers formidable flexibility. If, during a reread, I realized that a character's emotional transition was too abrupt between two events, all I had to do was update my PLAN.md to insert a new chapter.
Fed by the updated Git context, the AI generated that narrative bridge while respecting the continuity of the preceding and following files. Git versioning made it possible to test narrative pivots - story "branches" - and roll back without ever breaking the manuscript's integrity.
4. Multi-Model Review and Quality Control
One of the major challenges of AI-assisted writing is stylistic collapse. To address it, I set up a multi-model critical analysis workflow, where different AIs audited the text according to precise roles:
Gemini CLI (lore keeper): its role was to algorithmically verify that the chapter respected the bible and did not contradict the physical rules of my universe.
ChatGPT (dramatic analyst): it audited narrative rhythm, relational tension, and the characters' transformation arcs. It was the one that flagged when a conflict felt too artificial.
Mistral LeChat (stylistic editor): it provided a critical eye on fluidity, phrasing, and elegance of language.
Never relying on a single voice made it possible to obtain a text that was polished, critiqued, and reworked from every angle, while I remained the "showrunner" validating each commit in the repository.
5. Build Pipeline: From IDE to Physical Book
Since the novel was code, its publication had to be a software compilation. I created an automated script, build_book.sh.
From my terminal, running this script converted all the Markdown files in the chapters/ folder via Pandoc, applied a professional typographic layout with LaTeX, and generated the final deliverables in EPUB and PDF formats.
6. Transmedia Extension: Multimodality, Cover Art, and Vibe Coding the ARG
The universe of The Human Protocol lends itself perfectly to immersion, so I wanted to break the fourth wall. On page 175 of the physical book, a QR code invites readers to scan it and access the-human-protocol.com. This is not a showcase website. It is an in-universe clandestine archive node, the entry point to an Alternate Reality Game (ARG).
Here, multimodal AI brings all its power and creativity beyond text. In fact, the project's visual design, anchored consistently in the shared lore, began with the book cover.
Cover image generated with AI
The AI generated a strong visual aesthetic suited to the theme and universe of the novel: a pixelated silhouette against a geometric mountain background, crossed by a printed-circuit pattern.
This same visual identity then served as the foundation for the creation of the ARG website, entirely "vibe-coded" by Gemini CLI in a declarative way.
Homepage of the website the-human-protocol.com
To direct the developer AI, I provided it with the book PDF and the cover image as reference context, along with three strict Markdown specification files:
WHY.md(strategy): it defined the psychological goals: curiosity, exclusivity, and a feeling of belonging. It formally banned conventional marketing vocabulary ("Buy now," "Newsletter") in favor of an in-universe lexicon ("ACCESS," "SIGNAL," "FRICTION").WHAT.md(UX/UI): this file concretely translated the aesthetic of the book cover into an interface. It imposed a "Deep Void" blue-black background for depth, a "Protocol Cyan" accent color derived from the printed circuits and reserved for interactions, a technical typeface, and subtle animations to heighten immersion.HOW.md(technical architecture): the engineering brief imposed a modern stack to support server logic: Next.js 14 (App Router) in TypeScript, Tailwind CSS, and Prisma ORM for persistent database storage.
The site manages a true clearance mechanic, with authorization levels from 1 to 5. The reader progresses by solving puzzles based on the book, unlocking extended lore, hidden files, and access to a community of "Unlinked" readers.
ARG dashboard
The stack even includes an "Overseer Terminal" for administration: a secure dashboard used to audit user signals, adjust the campaign's global clearance level, and track in real time the number of scans of the physical QR code.
Conclusion: The Author-Architect Paradigm
Writing The Human Protocol proved to me that AI does not replace the writer: it reduces the barriers to production. The true value of a co-created work lies in the architectural rigor of its preparation.
By separating design (the lore), execution (the rules and prompts), and validation (multi-model review and Git), the creator becomes a true conductor.
Multimodality also opens the door to even broader transmedia horizons, such as a comic-book adaptation of the novel.
Excerpt from the comic book in progress
By applying similar engineering principles - namely, the explicit description of the drawing style in system prompts, as well as the creation of strict visual reference sheets, or character sheets, for the characters and technological elements - it becomes possible to extend the coherence and homogeneity of this universe into its graphic variations.
To go further technically, I am also considering creating specific AI "skills," or algorithmic capabilities, to further augment the design of the story by drawing on documented principles of dramaturgy and storytelling, and to refine the writing style by making it ever more explicit and controlled.
And ironically, it was by applying extreme software optimization processes that I was able to write a novel denouncing the loss of humanity in the face of algorithms.
About the Author
A writer and software architect who fully embraces his identity as a "Yogeek" - a point of balance between Yogi and Geek - Raphiki explores, across his work, the complex intersections between technology, consciousness, and humanity.
Writing under a pseudonym that reflects his dual nature as a playful seeker and an expert in cutting-edge technologies, he designs high-stakes thrillers that challenge our understanding of reality. His creative work often bridges the digital and the organic, drawing on his strong experience in open source innovation and emerging technologies.
When he is not deconstructing the fabric of dystopian realities in his manuscripts (or "vibe coding" them in his terminal), he can be found exploring the open source ecosystem or on a yoga mat.
Find his work, transmedia projects, and reflections at raphiki.github.io.
Top comments (0)