ReGen: Urban Ecosystem Architect

ReGen transforms urban decay into green ecosystems. AI turns your site photo into a photorealistic design and actionable construction plan, connecting you with local pros to build it.

Comment

🌱 ReGen — Urban Ecosystem Architect

Inspiration

Urbanization has quietly created a problem most people overlook—soil sealing.
Once you notice it, you start seeing it everywhere: empty parking spots, concrete courtyards, roadside strips of dead land. These spaces used to support life, but today they trap heat, worsen flooding, and eliminate biodiversity.

What stood out wasn’t just the scale of the problem—it was the gap between intent and action.

People want to fix these spaces. They want greener surroundings, cooler neighborhoods, and more biodiversity. But they don’t know:

  • where to start
  • what to plant
  • how much it costs
  • who to contact

ReGen was built to solve this exact moment of friction—the first step problem.

What We Built

ReGen is an AI-powered urban ecosystem architect that transforms a single photo into a complete, actionable ecological restoration plan.

Instead of just generating ideas, it delivers:

  • 📸 Multimodal site analysis
  • 🌿 Photorealistic future visualization
  • 🧠 Step-by-step execution roadmap
  • đź“Ť Real-world service recommendations

The system acts as an AI orchestrator, combining multiple capabilities into one seamless pipeline—from understanding a space to helping users actually transform it.

How We Built It

ReGen is built around Gemini 3 as a core reasoning engine, not just a text generator.

1. Multimodal Analysis

We use Gemini 3 Flash to process uploaded images and extract structured environmental insights using a strict JSON schema.

This includes:

  • Soil sealing estimation
  • Sunlight inference from shadows
  • Climate inference (e.g., hardiness zones)
  • Detection of ecological gaps

2. Generative Visualization

Using Gemini 2.5 Flash Image, we generate a realistic "after" image of the site.

The key constraint:

  • Preserve original perspective and scale
  • Ensure the output is believable and grounded

This helps users visualize change in a way that feels achievable.

3. Planning with Reasoning

We use Gemini 3 Flash with a thinking budget (~4096 tokens) to generate execution plans.

The model reasons through:

  • Budget constraints
  • Task dependencies
  • Material requirements
  • Time phasing

Example constraint logic:

$$ \text{Total Cost} \leq \text{User Budget} $$

$$ \text{Task}{n} \rightarrow \text{depends on} \rightarrow \text{Task}{n-1} $$

This transforms abstract ideas into a structured, week-by-week roadmap.

4. Real-World Grounding

To bridge the gap between plan and execution, we use Google Search grounding.

The system:

  • Finds local nurseries, landscapers, and services
  • Retrieves real contact details
  • Matches services to specific project phases

This ensures the plan is not just theoretical—it’s immediately actionable.

Challenges We Faced

1. From Vision to Execution

Early versions produced visually appealing outputs, but the plans were often impractical or inconsistent.
Introducing structured reasoning and thinking budgets significantly improved reliability.

2. Maintaining Realism in Image Generation

Generating beautiful images was easy.
Generating accurate, perspective-locked transformations was not.

We had to carefully constrain prompts and inputs to maintain spatial consistency.

3. Avoiding Hallucinated Plans

Without grounding and structured constraints, the system would:

  • Suggest unrealistic materials
  • Ignore sequencing logic
  • Misestimate costs

We solved this by combining:

  • Structured outputs
  • Reasoning constraints
  • External grounding

4. Bridging the “Last Mile”

Most AI tools stop at suggestions.
The real challenge was making ReGen actually usable in the real world.

Search grounding was critical to solving this.

What We Learned

  • AI is most powerful when orchestrated, not used in isolation
  • Multimodal inputs unlock entirely new categories of applications
  • Reasoning (not just generation) is essential for real-world tasks
  • The biggest barrier to impact is often not awareness—but starting

Impact & Vision

This is a problem visible in every city—which means the solution can scale everywhere.

ReGen enables:

  • Individuals to transform personal spaces
  • Communities to build micro-parks
  • Schools and campuses to restore unused land

Small, local actions can compound into large-scale ecological recovery.

ReGen is not just about generating plans—it’s about enabling action.

One photo → one plan → one restored space.

Share this project:

Updates