• Python 3

• Install Python dependencies:

pip install langchain
pip install openai

• Set up the OpenAI API key:

(Linux)

export OPENAI_API_KEY="<API_KEY>"

(Windows)

set OPENAI_API_KEY="<API_KEY>"

• Change the current directory to air-ai-backend

cd air-ai-backend

• Start the Web server

python server.py

• Node 16+

• Change the current directory to air-ai-frontend

cd air-ai-frontend

• Install package dependencies:

npm install

• Change the current directory to air-ai-frontend

cd air-ai-frontend

• Start the Web server

npm run dev

The frontend solution is pretty straight forward: get a JSON object from the backend server with cities as keys and their respective air quality index as values and render a speedometer for each city.

The backend solution consists of using the AI to evaluate the user query, extract the list of cities mentioned in the response and finally getting the air quality index for each city.

A user query can contain:

• a city that the user is interested in knowing its air quality

What is the air quality in Paris?

• a city that the user is not interested in knowing its air quality

What are the cities with the best air quality near Paris?

• no city names

What are the cities with the worst air quality?

So trying to extract the city names directly from the user query is not going to work.

• Use the browser Geolocation API for queries like:

What’s the current air quality? (where I am)

• Use a dedicated API for air quality index data as OpenAI API is not consistent for some cities and has no data for others.