Buildsmart AI

Inspiration

Planning permission is a frustrating process for homeowners and developers. The regulations are complex and confusing. Putting together an application is time-consuming. Getting a response can take ages, and applications are often rejected without clear reasons. We wanted to simplify the process and make it easier for people to successfully get planning permission.

What it does

The AI Planning Assistant guides users through creating a custom planning permission application tailored to their specific project.

How we built it

Claude2 - communicating with the regulation documents, applications, decisions and support documents LangChain - interacting with the language model Haystack Preprocessing - extracting text from PDF and DOCX, refine the text data MongoDB - storage and Atlas Vector Search for searching similar applications in an area MiniLM transformer (via HuggingFace) - building vector space from the texts Next.js - web application

Challenges we ran into

How can we simplify the process for the user (UX) for a highly complex journey To form the correct prompts

Accomplishments that we're proud of

We have a working prototype with real world data Found some customers at the hackathon The AI can generate customized planning applications to increase approval odds

What we learned

New prompting methods

What's next for Buildsmart AI

We plan to partner with more councils to access their data. After refining the product based on user feedback, we will distribute it widely and evaluate traction and impact. Our goal is to make planning permission easy and accessible for all

Built With

• claude
• haystack
• langchain
• mongodb
• nextjs