Hi HN!<p>I decided to get on the AI hype train and build my own version of an AI travel planner. Yes, it's pretty basic and there are existing ones like it out there, but this one's mine :)<p>Stack:
React, NextJS. I used this vercel template to get going quickly. A nice feature is that the API response is streamed in, so despite the 1-2m it takes to load the full response, users can already preview and won't click off right away.
<a href="https://vercel.com/templates/next.js/twitter-bio" rel="nofollow">https://vercel.com/templates/next.js/twitter-bio</a><p>There is no data fetching happening from external sources, it's two OpenAI api calls. To be more accurate, this could be replaced with serpapi + langchain + apify to fetch live data from the web, but I haven't needed to go down that rabbit hole yet. Who knows, maybe future models will have their weights updated daily with web content or someone will build a webscraperGPT api.<p>Now I'm waiting on API access to rail providers in Europe to build out the next set of features (rail + hotel booking). Joined Omio affiliate and in talks with Qunar.<p>To anticipate HN comments:
- you need JS to run it.
- is it expensive to run? About $1/day in API costs from GPT-3.5 turbo
- What's unique about it? How is this better than ChatGPT? So far, it's neither. But I've shifted focus to train journeys, and hope to show a cool update in a month or so.<p>If you have any tips, I would love to hear, especially if you've worked with any rail APIs.<p>PS I'm vlogging my journey from quitting my fulltime data science role to building startup(s)
<a href="https://youtu.be/FG63Y8l0e5A" rel="nofollow">https://youtu.be/FG63Y8l0e5A</a>