Turning written requirements into editable application flows

In spring and summer 2022, I used the OpenAI Playground alongside DALL·E, Midjourney, and Stable Diffusion to explore whether emerging generative models could help build interfaces. The experiments could sometimes produce front-end code or something that resembled a UI, but the composition was weak. Buttons appeared in arbitrary places, fields were poorly organized, and the results looked like interfaces made without an understanding of layout or interaction design.

The ChatGPT research preview in November 2022 changed what felt possible. After getting a simple website working, I built the first Semanttic alpha around a canvas. A prompt was decomposed into navigation, states, and multiple responsive mobile-web screens. The output was front-end only, with no editing or design-system support, but it established the core idea: generate a connected interface rather than a single image.

I began working on Semanttic full time in January 2023. The product then developed through the LAUNCH Accelerator, more than 200 founder-led interview and product-feedback sessions, the Techstars Miami spring 2024 cohort, and a Volvo CampX evaluation with more than 100 registered internal users. These programs matter here as product-research environments: each one exposed the workflow to more realistic requirements, users, and constraints.

The workflow was designed as a chain of useful product decisions. Each stage made the next one inspectable instead of hiding the process behind a single prompt.

Research also showed that the difficulty began before interface generation. Many people struggled to write a useful initial prompt, so I built ChatPRD.com as a companion tool for creating a structured product requirements document that could be brought into Semanttic. As model capabilities changed, we also built a router that assigned different tasks to the models that performed them best instead of tying the complete workflow to one model.

• A text-to-prototype workflow could support complex enterprise software rather than only landing pages.
• Product requirements, user stories, flow structure, visual composition, design systems, and code generation could be connected in one working environment.
• Early customer conversations and pilots exposed where generated output needed more control, transparency, and editability.
• Prompt support and task-level model routing were part of the product experience, not infrastructure concerns hidden from the design process.
• The strongest value came from compressing early product exploration while keeping teams able to inspect and change the result.

In April 2025, Semanttic's technology was acquired by a European consultancy, which uses it as a proprietary internal tool supporting its forward-deployed engineering services. I then joined Volvo to continue scaling AI-powered enterprise application development internally. I now also invest as an angel in early-stage AI startups.

My contribution spanned product development, product design, product strategy, research, interaction design, company building, sales, fundraising, hiring, and operations. The case remains centered on the product workflow, the evidence it generated, and what the team learned from putting it in front of customers.