BBA Publishing Platform
Inspiration
Companies and organizations are always coming up with new and useful information, whether it's about science, technology, or how they operate. But when it comes to sharing this knowledge, it's often all over the place and not very organized.
A source document can make its way through all sorts of channels - like email, chat, design tools, newsletters, websites, and social media platforms - without leaving a clear trail of where it's been.
- where the information originated;
- who reviewed it;
- what was approved;
- which version is canonical;
- how channel-specific adaptations were created;
- whether published content is still accurate.
The issue really came to the forefront when we were building the Axodus ecosystem. You see, Axodus generates a wide range of documents, including architecture papers, research reports, governance records, technical notes, release details, and official communications. To keep all these materials in check, we needed something more robust than just a social media scheduling tool or a run-of-the-mill content management system. We required a system that could handle the complexity and volume of our content, and that's when the problem became really apparent.
That need inspired the BBA Publishing Platform. The BBA Agency is part of Axodus and handles things like publishing, documentation, branding, and communication plans. They're currently building a platform for Axodus, but it's designed to be reusable so it can help other organizations later on. This platform will be a valuable tool for Axodus and other groups, making it easier for them to manage their publishing and communication needs. By creating a flexible and adaptable system, the BBA Agency is taking a forward-thinking approach that will benefit multiple parties in the long run.
Our guiding principle is:
The BBA Publishing Platform is not a marketing tool. It is infrastructure for managing the lifecycle of institutional knowledge.
What it does
The BBA Publishing Platform takes complicated information and turns it into something that's easy to control, review, and use again in a public setting.
The workflow is designed around the following lifecycle:
source knowledge
→ intake and classification
→ editorial planning
→ AI-assisted drafting
→ specialist review
→ institutional approval
→ canonical publication
→ channel-specific derivatives
→ analytics and maintenance
→ supersession or archival
The platform separates several concepts that are often incorrectly combined:
- source records;
- editorial drafts;
- approved publications;
- promotional excerpts;
- channel adaptations;
- internal or confidential information;
- archived and superseded versions.
So, when you share something, it's still possible to see where it originally came from, even if it's been changed to fit a website, newsletter, or other places like GitHub, LinkedIn, or a technical blog. This way, people can always tell who made it first, no matter where it ends up. The current prototype focuses on one bounded workflow: transforming a structured source document into a publication package containing:
- content classification;
- editorial brief;
- canonical draft;
- review recommendations;
- publication metadata;
- channel-specific summaries;
- an approval and audit record. People still need to give the final okay. AI tools can look at things, make drafts, check, change, and suggest, but they can't make the final decision on their own when it comes to science, technology, institutions, or publishing. They need a human to approve things.
How we built it
We designed the project as a modular, Git-based publishing workflow. Markdown is the main format used to share content between different systems. It uses structured metadata to keep track of important details like what type of content it is, who owns it, how sensitive the information is, if it's been reviewed, where it came from, what version it is, and if it's been published.
Git provides:
- version history;
- reviewable changes;
- traceability;
- rollback;
- immutable commit references;
- integration with automated validation workflows. GPT-5.6 is used to reason over complex source material and produce structured editorial outputs. Different AI-assisted roles can operate on the same content without sharing the same authority.
The initial agent model includes roles such as:
- Research Writer;
- Technical Writer;
- Editorial Reviewer;
- Scientific Reviewer;
- Brand Reviewer;
- Claims Reviewer;
- Distribution Manager.
Codex supported the implementation process by helping us:
- refine the project architecture;
- structure the content lifecycle;
- create reusable schemas and templates;
- design validation rules;
- implement workflow components;
- identify inconsistencies between documentation and implementation;
- accelerate iteration across the repository. Instead of asking one model to write and publish content autonomously, we designed a controlled sequence of specialized operations with explicit review boundaries.
Challenges we faced
Defining authority
The most difficult problem was not content generation. It was deciding who or what has authority at each stage. An AI-generated draft is not an approved publication. Editorial approval is not scientific validation. Brand review cannot override technical accuracy. A channel manager cannot silently change the meaning of an approved document. The platform therefore treats authority as an explicit part of the workflow.
Preventing semantic drift
When you start with one piece of information, you can create many different versions of it. But every time you change it, there's a chance that the original message, or how sure you are about it, or what it can't do, might get lost or altered in some way. We solved this problem by having one main version of the publication and connecting all the other versions back to it.
Separating public and private knowledge
Organizations need to publish useful information without exposing confidential roadmaps, security details, internal discussions, personal data, or unapproved claims. So, just because a document has been given the okay inside a company, it doesn't mean it's ready to be shared with the public. The decision to keep something private is separate from whether or not it's been approved internally.
Avoiding premature automation
Automating things to post on multiple sites at once is pretty simple. But doing it in a way that's responsible and makes sense is a whole different story. We learned that automation must include permissions, failure handling, audit records, rollback, and human approval. Otherwise, it only distributes mistakes more efficiently.
What we learned
The key takeaway was that publishing within institutions is basically about managing and overseeing how things are done.
Generating text is only one small part of the system. A reliable publishing platform must also manage:
- provenance;
- review responsibilities;
- disclosure;
- versioning;
- approvals;
- corrections;
- reuse;
- maintenance;
- archival.
We also found out that AI is really useful when it makes things easier for people to think and do their jobs, without taking away the important decisions that humans need to make. The best approach wasn't having one powerful system that did everything on its own. Instead, it was a team effort where different parts worked together, each doing their own job to help human reviewers make decisions. They would gather information, get everything ready, and then provide recommendations to the humans in charge. This way, the process was more organized and effective.
What is next
The next step is to finish a complete test of Axodus using a small group of real assets from an institution. The pilot will validate:
- source ingestion;
- structured classification;
- editorial transformation;
- specialist reviews;
- approval records;
- canonical publishing;
- controlled channel derivatives;
- analytics;
- correction and archival workflows.
After checking the operating model with Axodus, BBA Agency will look at the platform to see how well it works. They want to make sure it does what it's supposed to do.
- a managed institutional publishing service;
- an editorial consulting product;
- a reusable multi-client workflow;
- and later on, they might even develop a software platform to go along with it.
Our immediate objective is not maximum content volume or maximum audience reach. It is to prove that one complex source asset can move through a complete, defensible, and auditable publishing lifecycle.
Log in or sign up for Devpost to join the conversation.