Indian metro cities are home to millions of people. Most of the working class is not a resident of these cities. People spend a large amount of time searching for flats for rent over various platforms and are regularly frustrated by the overload of information. Voice bot solutions can bring ease in this search and make it smoother to search for houses.
What it does
FlatFinder can take data from various facebook groups containing flat rent, room rent, flat sharing posts and parse it through wit.ai model to extract structured data later this data is used to answer user queries. On FlatFinder user can ask about rental properties on basis of various properties like occupancy type (Single, double), size(in BHK), location etc. User can also create their own listing, using simple voice interface. E.g. "1 BHK semi furnished flat is available in Baner."
How we built it
- We trained 2 wit.ai bots, One for processing existing data to convert unstructured to structured format and other for processing user queries in three categories [help, create ad, find flat].
- Processed data is stored in a database for querying.
- Facebook messenger interface is used for user interaction.
- Python+Flask+Git+Heroku+Postgres is our tech stack.
Challenges we ran into
- 20 second limit of facebook messenger makes application harder to test.
- Slot filling have been challenging.
- Lack of tutorials around messenger and wit.ai.
Accomplishments that we're proud of
- Successful training of bots.
- Easy system for interaction built using voice commands.
What we learned
- Wit.ai is really powerful tool and can be used to solve a plethora of problems.
- Messenger integrated directly with groups can help generate more insights.
What's next for FlatFinder
- Get FlatFinder app approved by facebook; so that it can be integrated with groups to extract feed and generate private replies using wit.ai model.
- Improve wit.ai models with more data.
- Add visual elements for search.
- Improve search functionality.