GPT-recommend

Inspiration

Every time we go to a service business, we're always reading the reviews and figuring out what the experience is like to make sure we have a good time. We do this with restaurants, hair salons and especially with medical offices like doctors and dentists.

This takes a while. And it's mentally tiring to do this every time.

Why not let an LLM agent work on our behalf? We wanted AI to automate this process and give us well-informed recommendations that are personalized to our tastes and values.

What it does

We introduce a plugin and a web app. The web app allows the user to choose a business type and write a few words on what they're looking for. We send that query to GPT-3.5 to parse and give us two of the user's most desired features (that the business should embody). We then scrape the relevant queries off of Yelp, pass that data through a prompt template to GPT-3.5 which rates each business on the desired features mentioned above, extracts evidence for that rating (relevant text snippets from the reviews) and finally send the completed data payload back to us. Our colorful UI then displays these businesses, their ratings on the user's desired features and review highlights that are relevant to the user's query.

How we built it

React, Python, LangChain, OpenAI API.

Challenges we ran into

We started implementation at 10 pm yesterday.

Accomplishments that we're proud of

Finishing the end-to-end flow from user input to automation pipeline. Being functional after not sleeping.

What we learned

Building GPT plugins and apps on top of LLMs is easy and a great developer experience.

What's next for GPT-recommend

Deploy. Ship. Iterate.

Built With

• flask
• gpt
• react
• typescript
• vite