UrbanPlan

Inspiration

Techniques and systems put in place in ancient civilization to combat several environmental and urban obstacles have been some of the most innovative solutions. For example, Roman aqueducts moved water, Mesopotamian wind towers cooled buildings, and Persian qanats fought drought. We thought, "What if we could combine that old wisdom with today's AI, sensor data, and parametric modeling?" . That brought us to build UrbanPlan.

What it does

Our app is an AI driven urban design solution that leverages data from historical urbanization techniques combined with present technologies and resources available to create context aware smart urban solutions for the future.

How we built it

Frontend: Built with React /Next.js, the dashboard lets urban planners input site parameters such as location and concern to develop sustainability goals and instantly receive AI-generated adaptive design recommendations. Backend: A Python + Flask/FastAPI server handles all core logic, orchestrating requests between the frontend and the AI layer. A Node.js layer manages real-time communication and API routing, keeping response times snappy even during multi-step generation tasks. AI & Knowledge Layer: The intelligence behind UrbanPlan runs on OpenAI's GPT models paired with a LangChain RAG (Retrieval-Augmented Generation) pipeline. We built a custom knowledge base of historical climate-adaptive architectural techniques, bioclimatic design principles, and regional building strategies. When a user submits a site query, the RAG system retrieves the most relevant historical precedents and climate data, which GPT then synthesizes into context-aware, site-specific design recommendations grounding every suggestion in real architectural logic rather than pure generation. Data Pipeline: Historical case studies, passive cooling strategies, and climate records were chunked, embedded, and stored in a vector database, forming the long-term memory that makes UrbanPlan's suggestions historically informed rather than generically optimize

Challenges we ran into

• Front end back end integration: we had come challenges in having the front end able to fetch the backend data and being reflected correctly on the UI
• OpenAI Integration
• Debugging cross-service failures mid-hackathon was a real test of team coordination

Accomplishments that we're proud of

Built a working end-to-end RAG pipeline from raw site input to historically-informed design recommendations entirely within the hackathon window. The recommendations UrbanPlan generates aren't generic ; they're traceable back to real historical precedents, which gives them a credibility that pure generation can't match

What we learned

As we would consistently test our app, we would encounter several issues like crash outs, incorrectly rendered data, bad formatting. However, we were able to improve the app's ability to handle these issues.

What's next for UrbanPlan

1. Incorporate persistent memory so AI can utilize previous responses to generate better solutions
2. Enhanced context geographic awareness such that AI can tailor solutions to specific locations instead of general ones
3. Provide more parameters for users to customize the idea to be generated into an image and make it more interactive

Built With

• llm
• python
• rag
• react