Ozigi

This post tells the story of how these first five months of 2026 have gone for me.

The year started pretty much the same way the previous year ended: working a single contract job that wasn't bringing in substantial revenue.

I am a technical writer with over 5 years of experience, but I have always been a writer.

I had in mind to spread my tentacles and be more audacious with my job search and take boldness over best-fit (they say audacity is what gets you in a room, not your skills). I saw this job posting from a company I admire (name withheld on purpose). It felt like a holy grail moment to me because two things were at play with the job ad:

1. I had the audacity to apply because the pay jump was much higher than what my contract was paying.

2. I fit the job description perfectly. As a matter of fact, while applying, I already believed I had the job. The application was just a formality.

A constraint I noticed in the job ad was that they required the technical writer to post on X and LinkedIn every day, 2/3 posts a day. I was excited about the role, but I was least excited about this particular task. I kept thinking of what to do to overcome this particular challenge.

Then the idea of Ozigi was born. It didn't always start out with the name Ozigi. I created it to be an internal tool that I will pitch to the recruiters during the interview.

The idea was to create an engine that generates content. But generating high quantities of content wasn't enough; quality was a factor to consider as well.

So, I started tinkering with the prototype, and that is how the idea of the banned lexicon came to be. I knew a few words and patterns that were considered AI-slop, so I decided to just add them as system-level instructions for the engine to never generate content with those words.

But, I had so many other brilliant ideas:

1. I wanted to include direct posting so that I could post directly from the engine dashboard. I integrated LinkedIn, Slack, X, and Discord direct posting.

2. I wanted to be able to work through a whole week's worth of content and schedule it so that I could focus on other tasks. I initially used GitHub cron jobs to handle scheduled posts, but that failed at scale, so I moved on to Upstash.

3. I wanted to do more than just generate X and LinkedIn posts; I added Discord, Slack, and an email newsletter.

4. Generating a newsletter wasn't enough, though. How would I send it? Having to manually type out each email address to mail a newsletter to seemed like the kind of task no one should ever have to do. So, I integrated Zeptomail with Ozigi, and now users can send emails to their recipients directly from the Ozigi dashboard. List by manually typing out the email addresses once, or import with a CSV and have that list saved forever! Users on the team plan can make a generous offering of sending up to 500 emails at once. Higher limits for org and enterprise plans.

5. The banned lexicon idea was good, but new AI-speak words were coming out, and my lightweight constraint wasn't enough. I made a more robust constraint system and added a validator such that when the engine produces output A, it checks it against an AI-slop metric. If the score is high, it regenerates and has output B. It compares both outputs and shows the user the output with the least AI-slop resemblance. I go into more detail about this in this article.

6. I didn't want to send the output given to me by the AI engine directly. Even though I had eliminated almost all the AI slop from the content, it had no soul, and soul can only be given by a human. I enforced the 90:10 rule. Where AI does 90% of the work, generating a working first draft for you, and you give it the 10% that's uniquely you.

7. But having a 90:10 rule with the banned lexicon feature would make everybody sound alike, which defeated the purpose entirely. This is where the idea of the system personas came in. With the banned lexicon, a user's unique persona is served to the engine at the API level to ensure that it follows both rules before it even considers output generation. We also have a persona marketplace available to signed-in users, where they can choose from a well-curated list of personas to meet their different needs.

8. I create content, both for work and personally, at scale. And while I always add my human element to every piece of content, I needed to outsource a lot of the thinking and first draft work, so I built the Ozigi longform content generator. Available only to org and enterprise subscribers.

9. But having help generating content wasn't enough for me. For client articles, most of them came with a brief/outline that gave you the direction to follow. Personal articles, however? Those were another hassle. I had to come up with briefs first before I even thought of writing the content. This was manual work that was wasting time. So, I built the Ozigi technical brief generator under the long-form offering. It draws on the information you provided and sources from the web to create a well-informed, detailed brief for the user's needs. This was built with Gemini and Copilotkit.

10. But I'm not a builder; I'm a writer. Writers thrive better at writing. So, I started the Ozigi blog to keep up with my writing and reach a wider audience.

11. But content creation is hardly a one-man job, so I started a writers' community program. We have seen great progress so far. Published one article from the community, with two more in the pipeline for this week.

12. By this time, I already knew Ozigi had grown to be more than just an internal tool; it was something that needed to be out in the open. I started distribution.

13. I learned the biggest lesson of my life: you could find the cure to world hunger, but it wouldn't matter if no one knew about it or about you. I sadly was building in stealth throughout this process, hoping the goodness of my solution would rake in the customers. That was a big mistake. I am now trying to build more in public and form a consistent distribution pipeline.

14. Ozigi has done well so far with 50 registered users and a spike of 1800 visitors in total since it was publicly launched in April.

15. It has had to go through a series of landing page makeovers, but I think this current version suffices for the story I want to tell.

Building and distributing Ozigi is a job that I have done solo until now, and I am very proud of how far I have come.

The one thing that has reaffirmed my belief in this solution I built is that we are at the forefront of optimizing AI-generated content and making it standard and acceptable to everyone. Every other day, a new startup or model drops that creates more realistic or human-esque video/image, but textually generated content is left on the sidelines to suffer heavy criticism.

I tell people that AI-generated content is here to stay; our job is to create standards for it and provide guardrails on how it should be used.

I am Dumebi, and this is how I blow bubbles and fight crime!