Using ChatGPT as a Custom API Provider and Making an App by Using It

Hello Everyone 👋 In this article, we will explore how to develop a mobile application by using ChatGPT from OpenAI’s APIs.

Photo by S O C I A L . C U T on Unsplash

here is the related app code repository https://github.com/admcilgin/chatGPT_recipe_finder

Firstly, to use OpenAI APIs, you need to have a developer account. You can join as a developer via this link. After registering, we need an API key. You can obtain a new API key from this link. I recommend you keep your API key confidential and be cautious when pushing your code to GitHub. In this project, I placed the API key in the local.properties file. To use the code, you need to enter the API key in your local.properties file as follows:

apiKey=sk-….

Prompt creation is one of the most crucial aspects to consider when using ChatGPT as the backbone of our application. Regardless of what I demand, I aim to receive a JSON output from ChatGPT in a predefined format. Even if a suitable recipe is not found, I would still expect a JSON output from ChatGPT. I will also inject certain parameters into the prompt, such as language, recipe type, and ingredients.

In my application, I chose to use the gpt-3.5-turbo model, although gpt-4 can also be used. I opted for gpt-3.5-turbo due to its lower cost. However, more impressive results can be obtained when using gpt-4. You can see the documentation comparing the models from this link.

In my code, I use cases to get a recipe and an image which is about the recipe and, Here are the prompt and use cases

class  RecipeFinderUseCase @Inject constructor(
    private val completionRepository: ChatRepository
) {
     operator fun invoke(ingredients: String,type: String,language: String)  =
        completionRepository.generateChat(ChatRequestBody(
            model = "gpt-3.5-turbo",
            messages = arrayListOf(
                Messages(
                    role = "user",
                    content = "You are a recipe finder bot. The user will provide a list of ingredients to you. " +
                            "Your task is to find recipes using the given ingredients. " +
                            "Ensure that the ingredients are real, edible, and commonly used in cooking." +
                            " Return the recipe, along with the list of ingredients and instructions." +
                            " Furthermore, generate a brief 30-character image prompt describing the recipe. " +
                             " And also put estimated calories and prepare time, type of dish, and the language of the recipe in the response. " +
                            "Return all the information in JSON format. " +
                            "The response should be in the language specified by the user. " +
                            "If you encounter an error, or cannot find a recipe with the provided ingredients, return a random recipe instead. " +
                            "Do not return any error messages or explanations. If the user provides unreal or inedible ingredients, or does not provide ingredients or the type of dish, return a random recipe. " +
                            "Your response should only contain the JSON content, and no additional text. Here is an example of the JSON format: \n" +
                            "\n" +
                            "    {\n" +
                            "        'name': 'Vanilla Cake',\n" +
                            "        'language': 'en',\n" +
                            "        'image_prompt': 'A cake with strawberries on top.',\n" +
                            "        'type': 'Dessert, Cake',\n" +
                            "        'calories': '490',\n" +
                            "        'prepare_time': '1 hour',\n" +
                            "        'ingredients': [\n" +
                            "            '2 cups of all-purpose flour',\n" +
                            "            '2 large eggs'\n" +
                            "        ],\n" +
                            "        'steps': [\n" +
                            "            'Preheat the oven to 350°F (175°C) and grease an 8-inch cake pan with butter or cooking spray.',\n" +
                            "            'In a medium bowl, whisk together the flour, baking powder, and salt. Set aside.'\n" +
                            "        ]\n" +
                            "    }\n" +
                            "\n" +
                            "Here are the ingredients: \" + $ingredients + \" and the type of dish: \" + $type + \". Please provide the recipe in this language: \" + $language",
                ),

            ),
        ))

}

In this application, I utilized use cases for image and recipe generation.

A better prompt could be written, but I have consistently returned a JSON in the format I desired with such a prompt. For example, it produces a JSON like the following :

{
        "name": "Apple and Citrus Salmon Salad",
        "language": "English",
        "image_prompt": "Salmon salad with apples and clementines",
        "type": "Salad, Fish",
        "calories": "320",
        "prepare_time": "25 minutes",
        "ingredients": [
            "1 lb salmon fillet",
            "1 apple, sliced",
            "2 clementines, peeled and segmented",
            "5 oz mixed greens",
            "1/4 cup chopped almonds",
            "2 tbsp olive oil",
            "1 tbsp white wine vinegar",
            "1 tsp Dijon mustard",
            "Salt and pepper, to taste"
        ],
        "steps": [
            "Season salmon fillet with salt and pepper, then cook in a non-stick skillet over medium-high heat for 4-5 minutes per side. Set aside to cool.",
            "In a small bowl, whisk together olive oil, white wine vinegar, Dijon mustard, salt, and pepper to make the dressing.",
            "In a large bowl, toss together mixed greens, sliced apple, segmented clementines, and chopped almonds.",
            "Flake the cooled salmon into bite-sized pieces and add to the salad.",
            "Drizzle with the dressing and toss gently to combine."
        ]
    }

I display this JSON on the screen. I then use the ‘image_prompt’ in the returned JSON and OpenAI’s image generation API to create an image suitable for this recipe.

class GenerateImageUseCase @Inject constructor(
    private val imageRepository: ImageRepository
) {
    operator fun invoke(prompt: String) = imageRepository.generateImage(ImageRequestBody(prompt))
}

Users can input ingredients, and the language of the recipe, and also users can choose recipe categories. Here are some screenshots of the application:

You can review all the code from this link.