The problem was clear, and the stakes were high. Safety procedures save lives, so our solution had to do three things well:

1. Be accurate at scale - Mistakes in safety documents can have serious consequences. Our AI had to be precise and reliable, every time.

2. Have deep knowledge - Safety regulations are complex and always changing. Regular AI training wouldn’t be enough, so we built special knowledge into the system.

3. Be easy to use - Even though the platform saves time, it also had to be simple and fast. We didn’t want to add complexity just for the sake of new features.

We started with workshops to set priorities, plan the product, and decide what success looked like. After a few rounds, we chose an MVP that could be built in just two weeks.

On the technical side, we made some key choices:

We built a system that pulls in the right references, rules, and materials so the AI has the knowledge needed to write accurate procedures.

Instead of trying to create whole reports at once, we broke the process into smaller steps. Users could upload documents and references, which we stored securely. Our system pulled out the most relevant parts, summarized them, and then wrote each part of the procedure for the best accuracy.

We also used different AI models for different jobs, switching between more powerful and faster ones as needed.

⏱️ Complete safety procedures made in just 45–90 seconds.

For something this complex, we had to move fast. We did one or two demos each week, showing progress and collecting feedback right away.

To meet our goal, we skipped fancy design phases and built right from wireframes using Tailwind. We focused only on the basics like login and user profiles and rolled out updates for quick testing.

We finished the MVP on time, in just two weeks.

📈 Continuous Improvements

After launch, we kept making improvements. We let users regenerate just one paragraph at a time, added payments and subscriptions, built a chat-based way to enter data, and let people upload drafts and documents.