The leap to custom-built AI
The third stage in our AI journey was a turning point — the moment we transitioned from enhancing existing tools to building something entirely our own. Up until this point, our AI efforts had been built on top of hosted models, fine-tuned and shaped to our needs. But now, we were ready to construct an intelligence from the ground up, one that would be entirely under our control, tailored to our specifications, and capable of growing alongside our ambitions.
This new model had around 500 million parameters — a significant leap in complexity and capability. Training it was no small feat. We invested in high-performance local hardware and developed a training pipeline that allowed us to feed it vast amounts of carefully prepared data. The result was an AI that could handle much longer conversations, retain more context, and deliver nuanced answers that felt both natural and intelligent.
One of the defining features of this model was the introduction of two distinct modes. The first was a general chat mode, designed for everyday interactions. It was fast, responsive, and versatile, making it ideal for public-facing applications where users might ask anything from technical questions to casual prompts. The second mode was a training mode, built specifically for us. In this mode, we could provide feedback, corrections, and new examples, allowing the AI to refine itself over time.
This feedback-driven learning loop was transformative. It meant we could directly influence the AI’s behavior and knowledge without starting from scratch each time. Over weeks and months, it grew sharper, more accurate, and better aligned with our vision. In many ways, it was the first time our AI began to feel like a living project — one that evolved not just through code, but through collaboration between humans and machine.
Beyond its technical advancements, this model also represented a philosophical shift for Szymdows AI. We were no longer just users of someone else’s technology; we were creators of our own. It gave us the confidence and freedom to experiment without limitations, to tailor every detail to our needs, and to dream bigger about where our AI could go next.
This was the moment we truly felt ownership over our AI. It was hosted on our machines, trained on our data, and designed to meet our exact requirements. That sense of independence — combined with the flexibility to continuously improve — made it a foundational milestone in our development history.