Help for Agentic AI for Flight Booking Application | Learn AI Together | Page 1

astral sparrow Mar 10, 2025, 11:16 AM

#

We want to create an AI agent integrated into our in the making flight booking web app

Core Functionality :we want the agent to be able to search flight based on user input like, cheapest flight, earliest flight to certain city in the provided date etc. and include basic customer support and itinerary recommendations,
Integration: we will build a web app using next js, so it will be integrate in the web app first, may be mobile later,
Technology stack: please recommended an easy to use for our dev team that have no prior knowledge for LLM and Agentic AI, Lang chain, Lang graph?
User Interaction: Text based, a chat bot that intregate into the website and use tool calling to manipulate ui, like sending user to the flight search page and tracking the parameter from the conversation.

also please also recommended wether we should use commercial api ai (deepseek, llama, openAI, Claude, Gemini ETC) or locally host model using llm model that can be download?
and the main challenge is we mainly use thai language
we are currently testing it with 3B llma 3.2 model and only the flight search functionality, we are in the very early stage so any feed back will be really helpful, thank you so much

cold rock Mar 10, 2025, 3:10 PM

#

Go to Google cloud platform and create an account. You will have to pay a little to prove that you're not a bot, and in return you will get 300$ free credits. You can experiment with Gemini and Claude through Google Cloud.

Their portal has tutorials so you can follow that for learning.

thin nacelle Mar 10, 2025, 3:48 PM

#

This doesn't feel like a project that requires an "agent" outside of being the customer service front-end (ie "Jarvis, what're the cheapest flights to Montaco? I want to catch the Grand Prix with Pepper.")

You can try and have a zero shot classifier prompt for the input that determines which functions to call but the success rate may be a bit wonky. ie f"Given the following query: {user_input}, which of the following is most relevant: option 1, option 2, option 3, ... etc" then the output is piped/parsed over to the function which further deconstructs the user input into parameters. For instance, "I want to to go to Montaco" -> LLM chooses the book_trip() function -> LLM parses parameters "Montaco".

#

The problem with building these workflows is that you're relying on prompts to be able to extract all the necessary information, when you can build your own robust system to do that for you (no ~~agents~~ LLMs/prompts).

cold rock Mar 10, 2025, 3:51 PM

#

thin nacelle This doesn't feel like a project that requires an "agent" outside of being the c...

What you described is being marketed as "agent"

#

VertexAI has an agent builder kekw

thin nacelle Mar 10, 2025, 3:56 PM

#

cold rock What you described is being marketed as "agent"

I mean, classical definition of agent is exactly what I described. But whenever I hear people talk about agents, it usually ends up being "oh, I mean't LLM agents".

cold rock Mar 10, 2025, 3:56 PM

#

thin nacelle I mean, classical definition of agent is exactly what I described. But whenever ...

People who know the classical definition don't join this server elsa_pray