Hello, I’m Nikolai Taimanov.

I entered the Dubai Municipality AI Park Design Challenge and developed Al Safa Majlis Park, a proposal for a neighbourhood park in Al Safa 2.

Why I took on the challenge

Al Safa 2 has homes, a mosque, a medical centre, a school and a nursery around one small park, and I wanted that park to work for all of them on an ordinary Tuesday. I also wanted to test how much modern AI tools could help a person without formal architectural training develop a thoughtful proposal for a public space.

I trained as an electronics engineer, which is why I trust a measurement more than an impression, and I work in real estate investment and financial analysis, which is why the proposal carries a costed preliminary budget rather than an intention. I worked alone – no team and no consultants – bringing my experience of visiting parks with my family.

I began in early July 2026. I set the priorities, compared alternatives and decided what to keep.

What matters to me in a park

My family and I return often to Al Ittihad Park and Dubai Hills Park. I notice the running paths, the planting and the details that make us want to stay. I also find myself wishing for more benches in the shade and more exercise equipment for teenagers and adults.

Those observations shaped the proposal more than any reference project did. I was designing for people who come back several times a week, and for what makes an ordinary visit worth repeating.

The proposal in short

A majlis is the room where a household receives its guests. Al Safa Majlis Park brings that generosity into everyday public space: places to sit, play, walk, exercise and share water form six landscape rooms, with an open lawn left deliberately empty at the centre. A main loop of about 455 metres connects the rooms, so a visitor can choose their own pace and company. Shade is treated as infrastructure rather than decoration.

I could name five kinds of visitor for this park: families, walkers and runners, worshippers from the mosque next door, patients waiting with a carer, and teenagers. The design keeps them in one shared landscape instead of separating them within it. One of the checks I ran asked whether a wheelchair user could take a seat at the chess table.

How I developed the proposal

I started with research into the site and its surroundings. My method was to give the same assignment to several AI systems – Claude, Codex, ChatGPT, Gemini, Kimi, Grok, Perplexity, Manus and my own assistant Hermes – and compare their answers with each other, with sources and with the model. Where they disagreed, it usually meant I had not yet asked a precise enough question.

Image generation taught me an important lesson. Pictures from prompts alone could look convincing, but paths, trees and canopies changed between views. I needed a design whose layout and dimensions I could check.

So I reversed the order. I developed a 3D model in Blender with AI-assisted programming and measured the design from it using Python, NumPy and Shapely, with JRC PVGIS as the reference for solar energy. One check asked whether tree positions left enough room for people walking and running, counting the openings cut into the paving. It moved twenty-two trees.

The presentation illustrations were created with OpenAI’s image-generation tool in Codex, using views of that model as references, after earlier exploration with image models accessed through fal.ai. Generated images communicate atmosphere and use; plans, sections, dimensions and quantities are taken from the model, never from an image. The preliminary cost estimate is reported separately from those measured quantities, with low, base and high scenarios that show where the uncertainty sits. The one-minute animation is rendered from the same model, with an instrumental track from ElevenLabs Music that I chose and cut to the sequence.

I kept source files, alternatives and written decision records throughout, so I could always revisit what had changed and why.

What the proposal does not claim

The shade studies measure geometry, not thermal comfort. Keeping or relocating the existing trees needs an arborist’s survey. The five visitors are design scenarios, not interviews. This is concept and preliminary design, and those questions belong to the stage that follows.

What I learned

The project taught me to turn an idea into a question I could investigate: what would change, how would I check it, what should I keep? Learning to doubt an AI answer became part of using the tools.

Every decision in this proposal is mine. AI made my questions testable; it did not make the choices.

If you live in Al Safa 2, you know this park better than my research does. The five visitors above are my best guess at who comes and why, and I would rather be corrected than right about them.

My three-year-old son tested many playgrounds along the way. He remains a strict judge.

I am grateful for your time with this page. If it helps you see how the proposal was made and why I stand behind it, it has done its job.