AI Travel Agent

Implementing an AI Travel Agent using OpenAI's text API and LangChain framework to track information about future events with the help of duckduckgo search engine, wikipedia and other websites.

Project's Flowchart:

How it works:

• The API Works receiving a question query/prompt with instructions to an AI travel agent that returns an answer with a complete travel itinerary based on the informations provided by the user:

  • User could input the trip destination and dates and ask for a complete travel itinerary, events that will happen at the trip's time and flight tickets price, for example.
  • The AI agent then look up for the answer with the help of duckduckgo search engine, wikipedia, and other trip-related websites to ensure the most complete anwser possible.

How to set up AWS Cloud with a Docker Image and expose it to the Internet, Step by Step:

1. Create a `requirements.txt` file:

• 1.1. Make sure to create a requirements.txt file containing all the python packages you used in the project.

2. Create a image of your project with Docker:

• 2.1. Install Docker Desktop;
• 2.2. Create the Dockerfile;
• 2.3. Build the image:

  docker build --platform linux/x86_64 -t agent .

3. Setting up AWS ECR Repository:

• 3.1. Login at AWS Website;
• 3.2. Create a Private Repository at Amazon Elastic Container Registry (ECR) with the name travelagent, for example.

4. Setting up AWS CLI:

• 4.1. Install AWS Command Line Interface (CLI), copying the commands to your terminal, according to your Operational System;
• 4.2. Verify installation with the following commands:

  which aws
  aws --version
  

• 4.3. AWS Configure: requires AWS Access Key ID and AWS Secret Access Key.
  • To find those keys, go to AWS Identity and Access Management (IAM);
  • Then, find Users > your_user > Security credentials;
  • Create an Access Key, then copy paste your keys in the terminal, after typing the following command:

    aws configure
    

  • Lastly, confirm your region at AWS Home Console, next to your username, at the top right corner.

5. Push your API Image to your AWS ECR Repository via AWS CLI:

• 5.1. Head to the ECR Repository > View Push Commands.
  • Copy and paste the push commands into your terminal.
  • At this point, make sure to build, tag and push your image everytime there's a change to the code. Also, don't forget to deploy the new image to the Lambda function aswell.

6. Create Lambda Function:

• 6.1. Inside AWS Panel, search for Lambda;
• 6.2. Create a new function with a Container image:
  • Browse for the ECR Repository Image;
  • Select Architecture x86_64.
• 6.3. Edit function's configuration:
  • Increase Memory to 512 MB;
  • Increase timeout (e.g. 10 minutes).
• 6.4. Insert Environment Variable:
  • Inside the Lambda Function, go to Configuration > Environment variables;
  • fill the 'Key' field with OPENAI_API_KEY and the 'Value' field with the actual API Key.

7. Create Lambda Test:

• 7.1. Inside the Lambda function, go to Test:
  • Insert an Event name;
  • In this example, create a 'question' tag in the JSON Event with an example of an actual user prompt;
  • Save and Run the Test;
  • Check for execution status and details and see if corresponds to what's expected.

8. Exposing the application to the internet via AWS EC2 Load Balancers:

• 8.1. Inside AWS Panel, search for Elastic Compute Cloud (EC2);
• 8.2. At the left side Panel, look for Load Balancing > Load Balancers > Create Load Balancer > Application Load Balancer:
  • Scheme: Internet-facing;
  • Load balancer IP address type: IPv4;
  • Select the existing VPC and all Subnets;
  • Security groups: Default;
  • Protocol: HTTP, Port: 80, Create target group.
• 8.3. Creating a target group:
  • Target type: Lambda Function;
  • Provide a Target group name (e.g. target-travelagent);
  • Select the recently created Lambda Function, with Version set to $LATEST
• 8.4. Create Load Balancer (after adding the target group);
• 8.5. Edit Security configuration:
  • Head to Security > Open security group ID > Edit inbound rules:
    • Type: HTTP;
    • Port: 80;
    • Source: Custom -> 0.0.0.0/0.
• 8.6. Make sure to copy the DNS Name from the Load Balancer to call the API.

Results: Creating a call to the API.

try.curl file provides an example of a call to the API using curl. Here's the code:

  curl -X POST "http://api-travelagent-232686593.us-east-2.elb.amazonaws.com" \
  -H "Content-Type: application/json" \
  -d '{
      "question": "Vou viajar para Londres em agosto de 2024. Quero que faça um roteiro de viagem para mim com eventos que irão ocorrer na data da viagem e com o preço da passagem de São Paulo para Londres ida e volta em reais."
  }'

The output should be printed in your terminal.

EXTRA: Running the project through the frontend provided.

If you'd like to see the API working through a minimalist frontend, make sure to follow the steps:

1. After cloning this repository to your directory, install cors-anywhere as following:

   1.1. Clone cors-anywhere repository inside this repository's directory:

   git clone https://github.com/Rob--W/cors-anywhere.git
   

   1.2. Navigate to the directory and install dependencies with npm:

   cd cors-anywhere
   npm install
   

   1.3. Start the server at Port 8080:

   node server.js -p 8080
   

2. Now just open index.html file through Live Server VSCode Extension, or by just opening the file on your browser, and have fun 😄.

_{This project is part of the AI track at the 16th edition of the Next Level Week (NLW Journey) held by Rocketseat from 8-11 of July/2024.}