HostAI

Never answer the same question twice. Have an AI agent for your property on AirBnB to answer guest questions 24/7.

Comment

Inspiration

Our inspiration for HostAI comes from the afterhour "what's the wifi network id?" every traveller asks when booking with AirBnB. Guests need instant 24/7 support, but hosts are humans and can't always be there throughout the day. This delay leads to burnout from answering the same question multiple times. Even if the information is already present, it's common that it's buried deep with other booking details. We wanted to build a tool that gives guests hotel level service while giving hosts their time back.

What it does

HostAI provides rental hosts with a personalized 24/7 assistant that is uniquely tailored to a specific property, it obtains information about properties by proactively interviewing hosts using a list of common questions about check-in, wifi, parking, and amenities to build a data bank on common questions.

Guests are given a chat link to ask questions 24/7 to the AI agent, like where are the extra towels, how does the hot tub work, etc. and are able to get an instant answer from the AI. If the AI can't answer a complex or new question, it escalates to the host's "Training Inbox" The host then replies once, and the information is sent to the guest, whilst being permanently saved to the knowledge base for that specific property. The next time a different guest asks the same question, the host wouldn't have to open his app to respond. They would never have to answer, the same question twice.

How we built it

The basis for the webapp was full-stack with a RAG architecture, the: Frontend was built using React and Tailwind, and uses a similar color palette to AirBnB (pink/white accents) Backend is run by a FastAPI Python server that stores chat logic. Database for user data, and property information is stored via PostgreSQL LLM had used Anthropics Claude model for conversational reasoning, and it uses Pinecone and Langchain to store vector embeddings for each properties knowledge base.

Challenges we ran into

We faced a whole lot of challenges, personally it was exposure to a lot of new infrastructure I hadn't touched previously, specifically Langchain and using vector stores, another challenge was figuring out how to control the temperature of the responses by the LLM for the conversations. Occasionally it would not understand how to be comfortable with encountering missing information.

Accomplishments that we're proud of

I thought the proactive interview was a very nice touch, the process needs to feel intuitive, and as simple as possible to maintain the flow of how AirBnB hosts currently list their properties. I personally feel that it's a nice touch to only answer a question once, and let the agent take care of it by responding to future repeated questions.

What we learned

We learned that when having agentic AI deal with human interactions, its very important in some cases to have 'Human in the loop' feedback, I can't keep track of the number of times I had to nudge the AI in the right direction. Making the training process feel like a simple reply in an inbox is what makes this product valuable in my eyes.

What's next for HostAI

I would really like to have access to AirBnB's API to better integrate this, but at the moment HostAI stands as a proof of concept that there is value in giving hosts back time, doing this via AI is another step towards the right direction. It's easier on both the host, and the guest, when there's an agent to answer questions 24/7.

Built With

Share this project:

Updates