ADU Planner

ADU Planner_Pipeline

Inspiration

A lot of people in the US have the need to attach an accessory dwelling unit (ADU) in their backyard. The Accessory Dwelling Units (ADUs) market in the United States is estimated to be around $127.1 billion. The current process to find and plan for installation of an ADU is very time-consuming. We propose a brand new solution - GAI-powered ADU planner, to provide a personalized and user-friendly solution for generating an ADU plan for users, a solution to revolutionize the current ADU construction and to 5X speed up the process.

What it does

With our GAI-powered ADU planner, a user can get an ADU plan based on their property layout and their state’s ADU building policy by simply inputting their property address on our website. From the property address input, our website provides nearby ADU suppliers and floorplan options that align with the construction requirements. Users will be able to get a visual rendering according to their option.

How we built it

User inputs address,
- Pick options, e.g.: budget, layout (1b1b/studio), etc.
- (Use budget to inform sizing.)
Retrieve data from Google Maps
- Satellite imaging
- Dwelling location
- Lot size
Interpret building codes to identify buildable regions on lot
- Min. distance from the main dwelling, fire safety, utilities, etc.
Agent process:
- Collect sample layouts from local builders,
- Retrieve local building codes,
- Interpret local building codes (as design rules),
- Estimate price per square foot for construction
- Iteratively generate and test designs
Output top results from the agent process,
- 2D Render / Plotting
- 3D Render
App outputs potential placements for ADUs on their lot

Challenges we ran into

The image analysis capability of GPT-4V is not very stable, and we cannot get stable structured output to directly feed into the next function. Therefore, we need to additionally call a machine vision model to help us extract structured data. Moreover, because we need to calculate the accurate ADU constructible area based on the ADU policies of the user's residence, we have inserted a mathematical calculation module.

Accomplishments that we're proud of

We are proud that we have created an application that makes it much easier and faster to find and select compliant ADUs and will thereby increase ADU adoption and create less expensive housing for the many people in California and other locations across the United States who have a hard time affording housing.

What we learned

- Llama Parser + agent dramatically increased the information retrieval performance
- Front-end and back-end development and connection

What's next for ADU Planner

We will work with ADU vendors and set up agreements with ADU vendors so that we can get a commission. We will add a feature in which users can dynamically change interior and exterior wall coverings, floor coverings, cabinets and kitchen and bathroom fixtures using suggestions from Generative AI.