From pain to product

The Allomorphic story began in 2019 in Colorado. Jayson Go was a software architect at a Fortune 200 FinTech company. While there, he played domain architect roles as well as chaired several architecture review boards. Most folks knew him as JGo.

After seeing both sides of the governance process, as a submitter for his domains and as governance board chair, it became very clear to him that there was something broken. Many governance reviews were subjective and those reviews can take weeks or months. At that point, only the symptoms of the problem and superficial reasons for them were clear. However, the root cause was yet to be discovered.

Fast forward to 2022. By this time, JGo kept diving deeper into the governance problem while continuing to gain experience as an architect. By mid-2022, it became evident to him that the problem was related to how architects drew their boxes and lines diagrams. Those diagrams employed different names, used differing shapes and connectors with various semantics, and more importantly, the fact that architectures were defined by pixels. Thus, computing the delta between two architectural versions was generally impossible. This was the first aha moment: that architectures should not be defined by pixels, but instead with data.

By the October of 2022, JGo's startup idea was shaping up. He first named it ArchBook and subsequently registered archbook.io to formalize it. He socialized it with a few friends including those from other companies or FinTechs; they echoed the same pain-points which helped to validate the startup idea. It should be noted by the end of 2022, ChatGPT was released and was starting to gain traction from the general public. By early 2023, another aha moment occurred; if architecture's were captured as data (instead of pixels) then integrating AI would be more possible from summarization to chat. JGo reimagined ArchBook to include AI as a first-class citizen.

By 2023 and because of the infusion of AI capabilities into ArchBook, JGo decided to rename ArchBook to ArchMind inspired by Google's DeepMind, which solved many hard problems with AI. JGo also brought in an old friend, He Chen, into the company. Together, they created a C-Corp primarily due to their excitement that this startup had merit and investors may be interested. They decided that JGo would own a majority share of ArchMind.

JGo wrote all of the code for ArchBook. Initially, it was written with an Angular-Firebase-GCP stack but by 2024, he had re-written with a Flutter-Firebase-GCP stack. With Flutter, the code could render architectures to familiar boxes and lines diagrams, dark mode and vertical/horizontal orientations. For AI, he used Google Vertex AI specifically the PaLM2 models. In these earlier implementations, he used text-bison as a foundational model with training data, and ArchMind was able to provide the summarization and chat capabilities.

In the summer of 2025, JGo went to the Google Cloud Next event in Las Vegas as a speaker. He talked about the intersection between software architecture and AI, showcasing how the PaLM2 and Gemini models could be used with architectural data. In his talk, he demoed an early version of ArchMind. But, as successful as the talk was, the talk only gained superficial traction from some interested in learning more but not able to commit. Additionally, JGo met with three investors during the trip but all told him traction was the signal they needed to invest. And then, more bad news, albeit with a silver lining.

Despite the initial intent of capturing architectures as data, the design was centered around components. After all, this is what JGo, and most other architects, were used to. Take those boxes and lines, semantically define them then store them. This always bothered JGo because he knew that components were not the foundational elements; capabilities were. In fact, the earlier version of ArchMind allowed components to specify what capabilities they enabled, but it was always superficial. This was such an important aspect that JGo stopped investing more time into ArchMind.

Instead of bolting on more code to ArchMind, JGo decided to pivot. He needed to fully grasp his own views on modeling and this little thing called a Capability. To do this, he decided to write a white paper. This allowed him to not just capture his thoughts, but also validate every aspect of the idea. Of course, the white paper was not a "product" white paper. Instead, it was an academic one titled the Software Product Model. It modeled how software products are defined from inception to delivery. The parts he knew well were easily captured in the white paper, but as the topic of capabilities loomed closer, it wasn't so easy. JGo would write, then whiteboard, then write again. The white paper was an instrumental milestone for ArchMind and one that guided the product. JGo would pick up coding again, and once again rewriting it, but this time with the white paper to inform the features.

By the end of 2025, ArchMind had a LinkedIn page and a newsletter for the white paper. ArchMind was starting gain small attention from various vectors. There was another nagging problem that JGo had ignored. It was the name ArchMind. For many, when they read the name, they pronounce it with a soft-C like "starch" but the intended pronunciation is "ark". He noticed the pronunciation every where including an AI summary of the white paper. By this time, he had heard enough feedback or seen enough signal that this name is not sustainable that he once again renamed it. This time, Allomorphic. It took JGo weeks to arrive at this name due to other domain names being taken or just the myriad of options available. He used Gemini to get suggestions and eventually, it suggested Allomorphic. And after seeing that the domain allomorphic.com was available, JGo gave it serious consideration. It would take him another week to settle on the name and found that term allormorphic, though closely tied to linguistics, could be applied to software documentation. Generally speaking, allomorphic refers to some (linguistic) function that can have multiple forms for example if the function was pluralization, then for word the "boy", its plural form would be "boys" but for the word "man" it would be "men." When concept of the name was applied to the company, JGo considered that if the function was "software documentation", then it would have different forms depending on the need; for example, the same architecture could show business activities for a conceptual model while software components might be shown for a logical model. To JGo, it was a bit of a stretch where it would require continued explanations of how the name fits the mission of the company. By November 4, 2025, Allomorphic was born. For now, it's just a brand under ArchMind but soon it will replace it as the C-Corp.