During the course, I came up with the idea to extract valuable information from natural language. Online reviews seemed like a good place to start.
Unlike other platforms that pass reviews to an LLM and get it to output the sentiment, I came up with a formula for calculating how much each individual review comment affects the business's overall Google rating (see Reviews page of the demo).
The 'impact' values are then added up to tell business owners where they are doing well/poorly in terms of customer experience and suggests prioritised improvement actions (see Actions on Dashboard page) so that they know where to focus their attention.
So far, 25+ businesses are using the platform, mainly restaurants and food & beverage consultancies, but also cleaning services, bakeries.
I built the product myself, solo, it uses FastAPI backend, ReactJS frontend, Stripe for payments and OpenAI for the AI API.
A difficult part of building the app was creating the system message (prompt) that is used by the AI to extract the structured output from the reviews. This took many days of crafting, trial and error, not just to make it effective, but also cost effective for analysing hundreds of thousands of reviews, i.e. not just using the most expensive model.
Please give the platform a try, no signup required, and I have added 8 real businesses to the demo, with live data, which updates daily as customers leave reviews.
Let me know what you think?