Proposal for Yelp Fusion API Toolkit

[BaNg-W]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

We aim to integrate the Yelp Fusion API into a Langchain toolkit capable of:

• Finding the highest-rated businesses of a given business type (e.g., restaurant, cinema, shopping center) in a specified region.
• Finding the business details of a specified business, including business name, phone number, rating, price levels, and hours of operation.
• Getting prominent user reviews and/or Yelp review highlights for a business.
• Generating food venue recommendations based on a user’s location and preferred food and drinks.

Motivation

As more people use AI tools to help plan their outings, it is necessary for location-specific, up-to-date business and user information to be available to generate the best possible results. LangChain offers tools for querying business information such as Google Places; however, it particularly lacks in contexts where users need local service recommendations or business reviews. Yelp’s catalog of trusted user input and its wide range of established businesses can be a valuable asset to ensure quality responses to appropriate user queries.

Additionally, there has previously been a need for a working Yelp API integration, thus our tool would be the definitive solution for an easy and quick way to get a working Yelp chain: LangChain GitHub Issue #560

Proposal (If applicable)

We intend to create a toolkit enabling the use of Yelp’s Fusion REST API through query-specific tools utilizing a custom Yelp API wrapper, enabling LangChain agents to find appropriate businesses based on real-life data.

This would include additions such as:

• libs/community/langchain_community/agent_toolkits/yelp_fusion for the toolkit
• libs/community/langchain_community/tools/yelp_fusion for the tools
• libs/community/langchain_community/utilities/yelp_fusion.py for the API wrapper

Our team is a group of four friends, but we expect to complete the tool and its associated features in a quick time frame around mid-November.