Airbnb's Automation Platform v2 has been upgraded to support large language models (LLMs) for conversational AI applications.
The platform enables LLM development that assists support agents in efficiently providing resolutions and faster responses. It employs the use of LLMs in chatbot interfaces to enhance the natural conversation experience; these systems can interpret queries and gather nuanced information.
However, LLM-powered applications for large-scale experiences are still relatively new and aren't fully reliable for sensitive data validations, so the solution is to combine them with traditional workflows.
Beyond supporting LLM applications, the platform has been tailored to aid their development with features such as context management and guardrail frameworks and is fully integratable with common development tools.
Chain of Thought is a framework offered through LLM implementation, which allows reasoning for problem-solving. It uses an LLM as the reasoning engine, and tools as the ways for the LLM to interact with the world.
Context management ensures the LLM has all necessary and relevant information and can be customized for either statically declared context or named dynamic context retrievers.
LLMs can generate text and come with issues, like hallucinations and jailbreaks. This led to the creation of their guardrails framework, a safeguarding mechanism that monitors communications with the LLM to ensure they are helpful, relevant and ethical.
The platform aims to evolve with these technologies, expanding tool capabilities, and investigating LLM application simulation to provide gains for all AI practitioners.
Thanks to credits at the end of the post providing a team of engineers who worked together on the product, alongside an invitation to join the company if interested in work like this.
As of now, Automation Platform v2 is improving conversational AI at Airbnb, providing an enhanced customer support experience for clients.