On August 28th and 29th, 2018, teams from around the country will be meeting in San Francisco at NAR’s first iOi Hackathon to determine how artificial intelligence and machine learning impact the future of real estate. Teams can start building their ideas on August 1, and will present them as part of the iOi Summit on August 29th. It is imperative that all participants in the hackathon adhere to our Code of Conduct and frequently review the terms and conditions.
Hackathon Theme and Potential Project Ideas
Artificial intelligence and machine learning are already impacting real estate. From chatbots to natural language processing to predictive search, products are evolving. What is possible outside these realms? This hackathon aims to find new and novel approaches to how artificial intelligence and machine learning will impact real estate. Some example problems to explore:
- How can we improve the ways in which agents serve their clients?
- What are the opportunities for improving home search in the real estate industry?
- How can agents maximize their marketing opportunities with less effort and without opening themselves up to legal issues?
- How will data display be impacted by these technologies?
We aim to think deeply in this hackathon about these problems and how AI/ML can be applied to solve them.
Potential Projects (Solutions)
Realistic Agent Avatars
Advances in audio and video manipulation are improving every day. Videos of presidents, world leaders and celebrities are made to look like they are saying things that they aren’t actually saying using AI and ML software. Audio mimicry, where a person’s voice can be used to read text that is input, is being used in realistic sounding ways, with cadence and pronunciation being impersonated. Applications like Lyrebird use tweets to create realistic sounding audio of people reading them. Researchers at the University of Washington as well as German researchers are creating video avatars that can be manipulated by using direct audio input. Using b-roll video of the REALTOR and a copy of their synthesized voice, you could create a streamlined method for them to create promotional videos. What if these video and audio projects could be combined to allow for agents to create presentations on properties or CMAs on the fly with just text input? How could listing data be used to create enhanced presentations?
Use Computer Vision for Video or Image Parsing and Natural Language Generation
Parsing Images to Create Accurate Image Descriptions of Images for ADA Compliance
As part of ADA Compliance, all NAR members must make sure that they are updating the image descriptions for screen readers so that people with vision impairments can get correct information about listings. Doing so accurately and completely can take some time, especially when there could be anywhere from 10-20 images per listing. Using computer vision and natural language generation, these image descriptions could be auto-populated. This would allow a member to upload photos to their MLS, have the images parsed and for each image, description details that auto populate the HTML “alt” attribute.
Parsing Video for Creating Listing Details in Real Time
Another step that can be challenging for members is getting listing details entered into more than one MLS. One member could belong to several MLSs and need to create the property details for each one. In this idea, using video, you could create the property details via your phone’s camera. The NeuralTalk project in the Netherlands provides an example of how this could work. Moving along inside a property and filming as you move, the software could identify objects and features in a property and create a text file for use or automatically create a draft listing for the agent to review and edit. The video could also be used to generate images for the property listing.
Create a home price prediction model using only photos and listing details of the property and area
Automated Valuation Models (AVMs) have been around for years, but people have been concerned about their accuracy and usability. Some have been wildly inaccurate and caused issues with real estate professionals and their clients. What potential is there to create a home price prediction model that uses photos, both indoor and outdoor, to predict property price? Research papers on this possibility can be found here and here.