Building an AI-Powered Nutrition Label Analyzer

Gemini AI

CrewAI

OCR

In this tutorial, we'll build an AI-powered Nutrition Label Analyzer using Gemini for OCR and CrewAI for intelligent analysis. This project demonstrates how to combine multiple AI technologies to create a powerful tool for health-conscious consumers.

1. Setting Up the Environment

First, let's import the necessary libraries and set up our environment. We'll be using Gemini AI and CrewAI.

1import genai
2from crewai import Agent, Task, Crew
3from langchain_google_genai import ChatGoogleGenerativeAI
4
5# Set up Google API key (make sure to keep this secret!)
6genai.configure(api_key='YOUR_GOOGLE_API_KEY')
7
8# Initialize the language model
9llm = ChatGoogleGenerativeAI(model="gemini-pro", temperature=0.5)

2. Image Preparation Function

Next, we'll create a function to prepare the image for processing by Gemini.

1def prep_image(image_file):
2    """
3    Faz o upload da imagem para o Gemini e retorna o objeto do arquivo.
4    """
5    sample_file = genai.upload_file(
6        path=image_file,
7        display_name="Nutrition Label"
8    )
9    return sample_file

This function takes the path to an image file and uploads it to Gemini, returning a file object that we can use for further processing.

3. OCR Function

Now, let's create a function to perform OCR on the uploaded image.

1def extract_text_from_image(image_file, prompt):
2    """
3    Realiza o OCR da imagem e retorna o texto extraído.
4    """
5    model = genai.GenerativeModel(model_name="gemini-1.5-pro")
6    response = model.generate_content([image_file.uri, prompt])
7    return response.text

This function uses Gemini's GenerativeModel to extract text from the image. It takes the uploaded image file and a prompt to guide the OCR process.

4. Nutrition Label Analysis Function

Finally, let's create the main function that analyzes the nutrition label using CrewAI.

1def analyze_nutrition_label(label_text):
2    """
3    Processa o texto do rótulo nutricional com o agente e retorna a análise como dicionário.
4    """
5    nutritionist_agent = Agent(
6        role="Nutricionista",
7        goal="Analisar um rótulo nutricional e fornecer uma análise detalhada dos pontos positivos e negativos em tópicos.",
8        verbose=True,
9        memory=True,
10        backstory=(
11            "Você é um nutricionista experiente. Sua tarefa é analisar o texto fornecido de um rótulo nutricional, "
12            "identificar seus benefícios e potenciais riscos à saúde, e apresentar um resumo claro em tópicos."
13        ),
14        llm=llm,
15    )
16
17    nutrition_task = Task(
18        description=(
19            "Leia o texto do rótulo nutricional fornecido e analise seus pontos positivos e negativos. "
20            "Forneça uma lista clara e detalhada dos prós e contras relacionados aos valores nutricionais, ingredientes e outros aspectos importantes."
21        ),
22        expected_output="Uma lista enumerada de 1 a 5 em tópicos com os pontos positivos e negativos relacionadas ao produto.",
23        agent=nutritionist_agent,
24    )
25
26    crew = Crew(
27        agents=[nutritionist_agent],
28        tasks=[nutrition_task],
29        verbose=True
30    )
31
32    inputs_array = {'topic': label_text}
33    crew_output = crew.kickoff(inputs=inputs_array)
34
35    # Convertendo o resultado para algo serializável em JSON
36    result_text = str(crew_output)
37
38    # Processar o texto para separar pontos positivos e negativos
39    positive_start = "**Pontos Positivos:**"
40    negative_start = "**Pontos Negativos:**"
41
42    positives = ""
43    negatives = ""
44
45    if positive_start in result_text and negative_start in result_text:
46        positives = result_text.split(positive_start)[1].split(negative_start)[0].strip()
47        negatives = result_text.split(negative_start)[1].strip()
48
49    # Remover os asteriscos e formatar
50    positives = positives.replace("*", "").strip()
51    negatives = negatives.replace("*", "").strip()
52
53    result = {
54        "task_description": nutrition_task.description,
55        "positives": positives,
56        "negatives": negatives
57    }
58    return result

This function is the core of our Nutrition Label Analyzer. It uses CrewAI to create a nutritionist agent that analyzes the extracted text from the nutrition label. The agent provides a detailed analysis of the positive and negative aspects of the product based on its nutritional information.

5. Putting It All Together

Now that we have all our functions, let's see how we can use them together to analyze a nutrition label.

1# Example usage
2image_path = "path/to/your/nutrition_label_image.jpg"
3ocr_prompt = "Extract all text from this nutrition label image, including ingredient list and nutritional values."
4
5# Prepare the image
6prepared_image = prep_image(image_path)
7
8# Extract text from the image
9label_text = extract_text_from_image(prepared_image, ocr_prompt)
10
11# Analyze the nutrition label
12analysis_result = analyze_nutrition_label(label_text)
13
14print("Analysis Result:")
15print("Positives:", analysis_result["positives"])
16print("Negatives:", analysis_result["negatives"])

This example shows how to use all the functions we've created to analyze a nutrition label image. It prepares the image, extracts the text using OCR, and then analyzes the nutritional information using our AI nutritionist agent.

Conclusion

In this tutorial, we've built an AI-powered Nutrition Label Analyzer using Gemini for OCR and CrewAI for intelligent analysis. This project demonstrates how to combine multiple AI technologies to create a powerful tool for health-conscious consumers. By leveraging OCR and natural language processing, we can quickly extract and analyze nutritional information from product labels, providing users with valuable insights into their food choices.

This application has numerous potential use cases, from helping individuals with dietary restrictions to assisting in meal planning and nutrition education. As AI technologies continue to advance, we can expect even more sophisticated and accurate analyses in the future.