Learn how to become an Expert at Travel Planning with Tommy Travel AI

[@fastandsimpledevelopment]

People have asked what is the Tech Stack - Java 17, Spring Boot, AWS, Linux, Docker, MongoDB, Javascript, Python, Microservices

[@ExpertKNowledgeGroup]

Amazing concepts, love teh ChatGPT Integration

[@eprohoda]

Good night, Fast. super~you did good vlog. =)

[@fastandsimpledevelopment]

Thanks

[@lololoololdudusoejdhdjswkk347]

Great project. I was wondering, since you used an LLM for this, how did you integrate real time data and other things into it? Did you perhaps use some webscraper or real time API and then pushed into prompt? Sorry for the bombardment! Very very cool!

[@lololoololdudusoejdhdjswkk347]

I’m also interested in if you did do API calls, how did you guys change the query into the form of an API call? did you guys perhaps need to create dummy data in a database to knowledge embed or did you just create a prompt to configure API requests based on user input

[@fastandsimpledevelopment]

I created a context workflow at the beginning of the system so based on the context (one pass thru LLM) I then called external API's to collect that data (Like flight info for a date with prices) and then let teh LLM have access to that information. So yes I did use real time API for many things including Stock prices and history and Jira and Confluence as well as Slack integrations for the LLM

[@lololoololdudusoejdhdjswkk347]

Interesting, i wonder how much your AWS cost must be for hosting the LLM. I’m still a beginner, but If I were to try to recreate the structure I would assume you had a prompt set up to get context of the query , then made it so it would generate a desired response as well as a url(from context) to get the correct API you needed to call. Then I assume you would return this response. Is there perhaps a way for an LLM to call an API directly from the prompt?

[@lololoololdudusoejdhdjswkk347]

Look forward to more of your videos too

[@lololoololdudusoejdhdjswkk347]

Also would it be plausible to organize/ force the response from LLM to be a dictionary(url, response), so it’s much easier to classify the data and get what you want?