I Built a Platform Instead of Finishing My Book

I wanted to write a novel. I ended up building a platform instead. And I couldn't be happier.

It Started With a Python Script

If you read The Drakenhart Experiment, you know the backstory. I wanted to tell a story through a series of short-form videos shared on social media. AI-generated video, while absolutely amazing right now, still has a real consistency problem when trying to create the same characters over hundreds of short-form videos. I really didn't want to reduce the quality of my output just to fit the current industry's video capability, so I pivoted to telling my story through the written word instead, and started building a series of 5 fully written novels using AI assistance.

Sadly, I spent over a month struggling with AI to have it generate decent prose. As a developer, I applied my engineering skills to the problem and built a multi-tier system in Python to keep AI-assisted chapters coherent across a full novel - or 5 in my case. At its center was what I started calling the ChapterSpec - an AI-targeted specification that allows AI to know HOW to write my story without telling it what specifically to write. This allowed me to run the same generation multiple times while focusing on improving the prose it created instead.

It worked. Really well. The prose being generated around my story was very engaging and fun to read. It was at this point I began to realize that the spec-driven prose concept itself was actually fairly new. I couldn't find anyone else out there doing it in a similar way.

That's when I decided to share my tools with more people.

But... python doesn't share. It doesn't scale. You can't hand someone a terminal or a bunch of files and say "here, write your first book."

I had something I genuinely believed exceeded what the current tools in this space were offering. And zero other people could actually use it.

That bothered me more than I expected.

The process itself — going from blank concept to 150,000 words of prose worth reading — that's the part that's fun. That's the part that shouldn't require a software engineering degree to access.

So I decided to build a platform.

I Didn't Write a Line of Code

I'm a Principal Software Engineer with over thirty years' experience. When I decide to build something, I open an editor and start writing code or designing the architecture. However, building an entire platform by myself would easily take over a year and consume all the spare time I had. So before I started down this path, I made a deliberate decision: don't touch a single line of code unless there is literally no other way.

I vibe coded the entire Novelmint platform from day one.

This was my first real attempt to build a finished, production-ready app strictly through vibe coding. No scaffolding sessions. No "just this one thing." If I couldn't get there by describing it to an AI, I had to find a different angle. I applied my knowledge and skills to the prompt instead.

I am a solid AI convert now. The experience was phenomenal.

What surprised me wasn't that it could generate code. I expected that. What surprised me was the discipline it forced. Describing what you want means being precise about everything. My process was to describe the idea, expand it, come at it from multiple angles, build a proof of concept to validate the idea, and then finally apply it to the platform as a whole. The precision made the product better. I was able to focus entirely on the ideas, refine them, challenge them, test them, and build them through words alone.

What Was Actually Hard

Vibe coding wasn't the hard part.

Learning roughly a dozen third-party systems was.

ElevenLabs, Stripe, various automation tools, hosting, security audits, LLM APIs, storage, caching, etc. All of these I knew quite well through my day job, but the specific tools I used for Novelmint were new to me. Each had its own way of doing things that I had to learn and adapt to. Each integration required understanding not just their process, but the assumptions baked into it as well.

And in a vibe-coded solution, a lot of manual steps still remain. That's the part AI doesn't abstract away. I still had to understand the systems I was connecting. Thirty years of engineering instinct helped here, even when I wasn't writing the code.

Why I Thought the Community Side Was Solvable

From 2018 to 2022 I ran a social platform for independent musicians called Blackett Music. At its peak the site had ~4,000 members, ~2,000 artists, ~1,000 songs. I took two accounts on social media platforms from 0 to 50k followers over about a year and a half. I understood what was needed on the social front to build a community.

Sadly the business didn't survive the Twitter to X conversion, but it taught me one valuable thing I've held onto: build a good enough system, and you can build a community around it.

Most people see the marketplace problem — no readers, no authors, chicken and egg — and stop there. That history didn't stop me. It actually got me started. I'd done the community side before. I knew it was solvable if the product underneath it was worth building around.

This product is worth building a community around.

Where It Landed

Novelmint is live at novelmint.ai.

Studio is the author pipeline: sequence draft → review → prose → editorial → publish → promote. The spec system that started as Python scripts on my laptop is now a feature called ChapterSpec — a structured blueprint built before a word of prose is written. Authors steer at every stage. IP stays with the author. Always.

Commons is the reader marketplace. 5 cents per chapter, 70 percent goes to the author. First chapter of every series is free — the whole chapter, not a preview. A full 21-chapter book costs a reader 1.00 dollar. Typical production cost for an author: 50-80 dollars, get about 90 readers and you've made your money back, the rest is pure profit.

High-fidelity audio narration that sounds like a real human speaking. Support for translating into 15 languages (prose and audio). All noticeably superior to standard AI audio or translations because Novelmint can use the ChapterSpec to instruct the process. Author reliability scores baked in and Reader feedback that feeds directly back into the next chapter - all things real readers and real authors have wanted.

Novelmint's process stands out from others because it improves the quality of the reading or listening experience through the use of the ChapterSpec as a baseline for everything it produces. It produces more authentic output because it knows the intent behind what it is supposed to produce.

The catalog is small right now. That changes as authors publish. It's a growth curve.

I built this because I had something worth sharing and no good way to share it.

That is still the whole reason.

Novelmint is a TeamLesh LLC product — novelmint.ai