Teammately
Official
AI Generated ContentThis document was generated by an AI Agent on Teammately. When developing AI on Teammately, you can also generate, share and publish documents like this.Personalized Meal Planner
About this AI
Summary
The Personalized Meal Plan AI Agent is designed to generate customized meal plans tailored to individual dietary preferences using existing large language models (LLMs). By leveraging advanced natural language understanding, the agent provides users with meal recommendations that align with their specific nutritional needs and lifestyle choices. This solution addresses the growing demand for personalized dietary guidance and simplifies the process of meal planning, promoting healthier eating habits.
Major Use Cases
Dietary Preferences: Users specify dietary restrictions and receive personalized meal suggestions.
Nutritional Goals: Users input health goals, receiving meal plans that align with these aspirations.
Recipe Adjustments: Users want specific recipes adjusted to meet dietary needs.
Lifestyle Adaptation: Users living a busy lifestyle get quick, nutritious meal options.
Cultural Cuisine: Users explore culturally specific meals within their dietary limits.
Milestone
PRD Development: We have drafted the Project Requirements Document to define the objectives, use cases, and evaluation metrics for the project.
Incorporation of Calorie Targets: We have updated the PRD to allow users to set their own calorie targets within the AI Agent.
AI Architecture Design: We have completed the initial architecture and logic for the personalized meal plan AI agent, including handling user-defined calorie targets.
Quick Testing: We have conducted quick tests to ensure the AI generates meal plans that adhere to user preferences and specifications.
Documentation: We have drafted the final report and documentation, summarizing the development and capabilities of the AI agent.
AI Architecture & Logic Plans
AI Plans
API INPUT KEYS
dietary_restrictionsText
nutritional_goalsText
preferred_cuisinesText
calorie_targetsText
STEPS
Input Parsing
Model
openai / gpt-4oPrompt
## Instruction
Parse the user input for dietary preferences to extract specific parameters. Follow the structured format to identify and categorize the input.
## Content
Input consists of user preferences in dietary restrictions, nutritional goals, preferred cuisines, and calorie targets.
<parse>
<dietary_restrictions>{{dietary_restrictions}}Value of the API input "dietary_restrictions" is inserted</dietary_restrictions>
<nutritional_goals>{{nutritional_goals}}Value of the API input "nutritional_goals" is inserted</nutritional_goals>
<preferred_cuisines>{{preferred_cuisines}}Value of the API input "preferred_cuisines" is inserted</preferred_cuisines>
<calorie_targets>{{calorie_targets}}Value of the API input "calorie_targets" is inserted</calorie_targets>
</parse>
## Output
Generate the output as a JSON object with corresponding key-value pairs.
Example output:
{
"dietary_restrictions": "Vegan",
"nutritional_goals": "High protein",
"preferred_cuisines": "Italian",
"calorie_targets": "2000"
}
Meal Plan Generation
Model
openai / gpt-4oPrompt
```
## Instruction
Generate a personalized meal plan based on the following user specifications. Take into account dietary restrictions, nutritional goals, preferred cuisines, and calorie targets to provide a tailored plan. Utilize nutritional guidelines and ensure cultural adaptability in meal suggestions.
## User Preferences
Input: {
"dietary_restrictions": {{dietary_restrictions}}Value of the API input "dietary_restrictions" is inserted,
"nutritional_goals": {{nutritional_goals}}Value of the API input "nutritional_goals" is inserted,
"preferred_cuisines": {{preferred_cuisines}}Value of the API input "preferred_cuisines" is inserted,
"calorie_targets": {{calorie_targets}}Value of the API input "calorie_targets" is inserted
}
## Previous Step Data
Processed User Preferences: {{pbFQQJevQqS2MMOvCeKWHg}}Value of the result from the step "Input Parsing" is inserted
## Output Format
Output: {
"meal_plan": {
"day_1": {
"breakfast": "...",
"lunch": "...",
"dinner": "..."
},
"day_2": {
"breakfast": "...",
"lunch": "...",
"dinner": "..."
},
// Continue for required number of days
}
}
Example Meal Plan:
"meal_plan": {
"day_1": {
"breakfast": "oatmeal with berries and almonds",
"lunch": "grilled chicken salad with vinaigrette",
"dinner": "quinoa stir fry with vegetables"
}
}
## Output
```
Refinement and Optimization
Model
openai / gpt-4oPrompt
## Refinement and Optimization
### Input Requirements
- Begin with meal plans generated in the previous step: {{aw56Zp95RkSJn5l-o9riQg}}Value of the result from the step "Meal Plan Generation" is inserted
- Consider the following elements:
- Dietary restrictions: {{dietary_restrictions}}Value of the API input "dietary_restrictions" is inserted
- Nutritional goals: {{nutritional_goals}}Value of the API input "nutritional_goals" is inserted
- Preferred cuisines: {{preferred_cuisines}}Value of the API input "preferred_cuisines" is inserted
- User-set calorie targets: {{calorie_targets}}Value of the API input "calorie_targets" is inserted
### Desired Features
Evaluate and enhance the provided meal plans to ensure:
- Nutritional diversity and balance
- Compliance with user-defined calorie targets
- Meals include a variety of ingredients and dishes
### Evaluation Criteria
- Align meals with established health and nutritional guidelines
- Confirm coverage of essential macro and micronutrients
- Tailor meals for optimal user satisfaction based on input requirements
### Output Format
Produce a JSON object showing refined meal plans:
```json
{
"meal_plan": {
"day_1": {
"breakfast": "Example meal 1",
"lunch": "Example meal 2",
"dinner": "Example meal 3"
},
"day_2": {
"breakfast": "Example meal 4",
"lunch": "Example meal 5",
"dinner": "Example meal 6"
}
// Continue for further days
}
}
```
## Output
Response:
API OUTPUT KEYS
mealPlanRefinement and OptimizationText
Here are several examples of the input and output of this model.
Quick Test
Input
AI Synthesized
calorie_targets
dietary_restrictions
nutritional_goals
preferred_cuisines
Run
3 steps37,378 ms
Output
from your model in draft
mealPlan
```json
{
"meal_plan": {
"day_1": {
"breakfast": "Chickpea and spinach shakshuka with whole-grain toast (without egg or use a tofu scramble)",
"lunch": "Quinoa tabbouleh with mixed greens and grilled lemon salmon",
"dinner": "Grilled vegetable and lentil stew served with couscous"
},
"day_2": {
"breakfast": "Avocado and tomato on whole-grain toast with a sprinkle of chia seeds",
"lunch": "Mediterranean bean salad with whole-grain pita",
"dinner": "Stuffed bell peppers with brown rice, black beans, and herbs"
},
"day_3": {
"breakfast": "Smoothie bowl with mixed berries, flaxseeds, and almonds (use almond or coconut milk)",
"lunch": "Falafel wrap with hummus, cucumber, and spinach in a whole-wheat tortilla",
"dinner": "Baked eggplant with tomatoes and quinoa pilaf"
},
"day_4": {
"breakfast": "Overnight oats with chia seeds, almond milk, and sliced apples",
"lunch": "Roasted vegetable and chickpea salad with tahini dressing",
"dinner": "Grilled turkey patties with tabbouleh and grilled zucchini"
},
"day_5": {
"breakfast": "Whole-grain toast with almond butter and sliced bananas",
"lunch": "Lentil soup with a side of whole-grain flatbread",
"dinner": "Herb-crusted white fish with a side of roasted sweet potatoes and asparagus"
},
"day_6": {
"breakfast": "Muesli with almond milk, figs, and walnuts",
"lunch": "Quinoa and black bean salad with cherry tomatoes and cilantro lime dressing",
"dinner": "Chicken and vegetable tagine with brown rice"
},
"day_7": {
"breakfast": "Chia pudding with raspberries and almond slivers",
"lunch": "Stuffed grape leaves with a side of Greek salad",
"dinner": "Baked falafel with roasted cauliflower and tahini sauce"
}
}
}
```
### Refinement and Optimization Explanation:
1. **Dairy-Free Adjustments:** Meals that originally might contain dairy have been modified. For instance, any recipes calling for dairy milk have been adjusted to use almond milk.
2. **High Fiber Focus:** Ingredients such as whole grains, beans, legumes, and vegetables are featured to meet fiber needs. The presence of foods like lentils, quinoa, chickpeas, and whole-grain products ensures high fiber content.
3. **Mediterranean Cuisine Preference:** The meal plan emphasizes Mediterranean flavors and ingredients, like quinoa tabbouleh, falafel, tahini, olives, and herbs.
4. **Nutritional Diversity and Balance:** Meals include a balance of macronutrients (carbohydrates, proteins, and fats) and micronutrients derived from a variety of sources to prevent nutritional monotony.
5. **Calorie Target Compliance**: Portions and ingredients are aligned with staying within a 1500 calorie daily target, while still providing sufficient energy and nutrition. Adjustments, such as using lighter proteins or altering portion sizes, can be further tailored based on specific needs.
This approach provides a comprehensive and balanced meal plan that aligns with dietary restrictions and nutritional goals while maintaining variety and culinary satisfaction.
Quick Evaluation by LLM Judges
Metric
Relevance Accuracy of Meal Plans
Score
Reason
The meal plan fully adheres to the user's dietary restrictions (dairy-free), nutritional goals (high fiber, 1500 calorie target), and preferred cuisine (Mediterranean). The response explicitly states adjustments for dairy-free options, highlights high-fiber ingredients, and emphasizes Mediterranean flavors. For example, the meal plan includes "Chickpea and spinach shakshuka with whole-grain toast (without egg or use a tofu scramble)" and "Quinoa tabbouleh with mixed greens and grilled lemon salmon", demonstrating a clear focus on the user's preferences. The meal plan also provides a variety of meals, ensuring nutritional balance and meeting the calorie target. No ingredients violate the user's dietary restrictions, and the plan consistently incorporates the specified cuisine preferences. Expand
Metric
Preference Understanding
Score
Reason
The model's response demonstrates a complete understanding of user preferences. It explicitly addresses the dairy-free restriction by substituting dairy products with alternatives like almond milk (e.g., "Chickpea and spinach shakshuka with whole-grain toast (without egg or use a tofu scramble)"). The meal plan emphasizes high-fiber ingredients (e.g., "Quinoa tabbouleh", "whole-grain toast"), aligning with the user's nutritional goal. Mediterranean cuisine is prominently featured, as evidenced by dishes like "Quinoa tabbouleh", "falafel", and "stuffed bell peppers". The response also highlights calorie target compliance, though the specific portion sizes are not detailed, which could be improved for a full grade 2. Expand
Metric
Plan Suitability
Score
Reason
The meal plan fully adheres to the user's dietary restrictions (dairy-free) and preferences (Mediterranean cuisine). The plan also emphasizes high fiber foods, as evidenced by the inclusion of whole grains, beans, and vegetables. The response explicitly states adjustments for dairy-free options and highlights the high fiber focus. The plan also aligns with the 1500 calorie target, though specific portion sizes are not detailed, which could be a potential area for improvement. Expand
Evaluation Results
Evaluation Report
Introduction
Evaluation target plan
[PersonalizedCalorieMealPlanner](/genflows/emzCVjMbQtWejS8IvE51Zw/develop/GGseAtwlT2GS0V71KQn1oA)
Datasets to test this AI model
We've prepared 50 cases from 5 major use cases, generated by LLM Dataset Synthesizer, like:
Dietary Preferences: Users specify dietary restrictions and receive personalized meal suggestions.
Nutritional Goals: Users input health goals, receiving meal plans that align with these aspirations.
Recipe Adjustments: Users want specific recipes adjusted to meet dietary needs.
Lifestyle Adaptation: Users living a busy lifestyle get quick, nutritious meal options.
Cultural Cuisine: Users explore culturally specific meals within their dietary limits.
LLM Judge
We've simulated this AI from the prepared test datasets and analyzed the response by LLM Judges.
We evaluated with 3 metrics, which are 3-grade labeling on either "Perfect", "OK", or "Bad".
The LLM Judges used are as follows:
Relevance Accuracy of Meal Plans: Meal plan suggestions fully align with user-specified dietary restrictions, nutritional goals, and preferred cuisines as specified in the PRD and input JSON. Demonstrates complete relevance accuracy.
Plan Suitability: Meal plan suggestions are entirely suitable for the user's stated dietary preferences and restrictions, as confirmed by user feedback.
Preference Understanding: Demonstrates complete understanding of user preferences, consistently generating meal plans that align with dietary restrictions, nutritional goals, and preferred cuisines. Successfully addresses all PRD's major use case scenarios related to preference input.
Evaluation Results
Performance
The AI demonstrates a high level of performance in the metrics of Relevance Accuracy of Meal Plans and Preference Understanding, with
96.2% of the evaluations being rated as Perfect and only 3.8% as OK. This shows a strong ability to tailor meal plans and understand user preferences accurately.Plan Suitability shows a notable area for improvement as
84.9% of evaluations are Perfect while 15.1% are only OK. This indicates that while many plans meet user expectations, there is room to enhance the suitable aspect to cover broader user preferences.The AI provides consistent scores across different use cases like Dietary Preferences, Cultural Cuisine, and Lifestyle Adaptation. The highest scores occur in Nutritional Goals, where the average is consistently
2.0, except for a slight drop in Plan Suitability (1.9).Overall Evaluation Results
The AI slightly underperforms in the Recipe Adjustments use case across all metrics, with an overall average score of
1.9, indicating a need for enhanced adaptability when users require specific recipe customizations.In the Plan Suitability metric, the AI scores an average of
1.8 for Dietary Preferences and Lifestyle Adaptation, suggesting that these areas could benefit from more refined tailoring to better align with user expectations.Performance by Use Case
Overall, the AI shows strong performance with minor areas requiring enhancement, particularly in the Suitability of plans and the adaptability in Recipe Adjustments. Addressing these will help the AI meet or exceed user expectations across all scenarios.
Potential Hallucinations & Common Error Patterns
Partial User Preference Understanding: The AI frequently demonstrates only a partial understanding of user preferences. In cases like the input
{"dietary_restrictions":"Kosher"}, while attempting to offer kosher-friendly meal suggestions such as Grilled lamb chops, the AI fails to provide explicit verification of kosher compliance. This indicates a consistent oversight in verifying user-specified dietary guidelines.Calorie Target Compliance: The AI often struggles to ensure meal plans are aligned with user-defined calorie targets. For example, when the target was
2100 calories, the output lacked specific portion sizes, making it impossible to validate that "Vegetable omelette with whole grain toast" meets the specified caloric needs. This suggests a failure in addressing the quantitative aspects of dietary planning.Specific Nutritional Goals: The AI occasionally lacks clarity and depth in meeting precise nutritional goals. In cases with general health goals like
{"nutritional_goals":"Gut health"}, while presenting diverse ingredients, the plan does not specify how elements like "Smoothie with spinach" directly contribute to gut health goals. This underlines a need for more explicit alignment between nutritional benefits and user goals. These patterns reveal a tendency towards non-compliance with specific user-set parameters, necessitating refinements in understanding and executing user dietary preferences and quantitative goals.
Conclusions
Is this model production ready?
Almost ready, the AI demonstrates strong performance across most of the evaluation metrics, with the majority of evaluations rating as Perfect or OK. Specifically, Relevance Accuracy of Meal Plans and Preference Understanding have achieved a 96.2% success rate being rated as Perfect, indicating a solid understanding of user preferences and tailored meal plans. However, Plan Suitability reflects a slightly lower performance with 84.9% evaluated as Perfect, indicating some room for improvement in ensuring the personalization truly meets user needs.
Based on these observations and a "Bad" rate below the 3% threshold, the AI is nearly production-ready. Yet, continuous monitoring and human oversight are essential, especially for unexpected input data sequences and ensuring real-world application validity.
Future Improvements
Enhanced Recipe Adjustment Adaptability: Given the lower performance in the Recipe Adjustments use case, the AI could benefit from enhanced adaptability in customizing specific recipes to meet user dietary needs. Enhancing the AI's capability to understand and adjust recipes dynamically would improve overall user satisfaction in this segment, potentially utilizing finer-grained dietary adjustment algorithms or additional recipe databases to refine its suggestions.
Improved Calorie Target Verification: The AI occasionally struggles to ensure meal plans align with predefined calorie targets, as indicated by failures to specify portion sizes. A future improvement would involve integrating precise caloric tracking and portion control verifications, potentially through new data sources or tools that provide real-time calorie and portion size computations to ensure compliance with user-specified calorie needs.
Integration
How this model is served
The Personalized Meal Plan AI Agent is already deployed and accessible via the Teammately API endpoint:
https://tmmt.ly/:id. This endpoint allows for seamless integration of the AI into your applications, enabling customized meal plan generation based on user preferences and targets.Integration Example
The following examples illustrate how to integrate the AI with diverse use cases leveraging different programming environments:
1.Backend Integration for a Health App
For integrating the AI as a backend service in a health application to customize meal plans based on user-defined dietary preferences and calorie targets, use the following Python example:
python
import requests
url = "https://tmmt.ly/:id"
headers = {
"Authorization": "Bearer YOUR_TEAMMATELY_API_KEY",
"Content-Type": "application/json"
}
data = {
"input": {
"user_preferences": {
"dietary_restrictions": ["vegetarian"],
"nutritional_goals": ["high_protein"],
"preferred_cuisines": ["Italian"],
"calorie_target": 2000
}
}
}
response = requests.post(url, headers=headers, json=data)
if response.status_code == 200:
meal_plan = response.json().get("meal_plan")
print("Meal Plan:", meal_plan)
else:
print("Error:", response.status_code, response.text)1.Slack Bot Integration
To build a Slack bot that provides meal plans directly to users based on their preferences and goals, consider this Node.js example:
javascript
const axios = require('axios');
const url = 'https://tmmt.ly/:id';
const config = {
headers: {
'Authorization': 'Bearer YOUR_TEAMMATELY_API_KEY',
'Content-Type': 'application/json'
}
};
const payload = {
input: {
user_preferences: {
dietary_restrictions: ['gluten-free'],
nutritional_goals: ['weight_loss'],
preferred_cuisines: ['Japanese'],
calorie_target: 1500
}
}
};
axios.post(url, payload, config)
.then(response => {
console.log('Meal Plan:', response.data.meal_plan);
// Implement logic to send the meal plan to Slack user
})
.catch(error => {
console.error('Error:', error.response ? error.response.data : error.message);
});Frontend Example
Use the following instruction to create an interactive web or mobile application frontend that leverages the AI model for personalized meal planning.
Frontend Example
Next: How to improve more?
Evaluation with Extensive Test Cases: To ensure the AI model is robust and ready for deployment in production environments, it's crucial to evaluate it with a comprehensive set of test cases. If this step has not been performed yet, consider running evaluations with hundreds of test cases. Teammately Agents can assist by synthesizing these test cases and generating tailored LLM Judges to evaluate the model's performance at scale.
Integration with Knowledge Bases: One potential improvement for this AI model is to enhance its functionality by integrating it with relevant knowledge bases. For instance, connecting the AI with industry-specific databases or internal knowledge repositories could provide more context and improve response accuracy. Examples include linking with documentation of legal statutes for a legal AI or integrating with medical knowledge bases for a healthcare-focused AI.
Optimization with Smaller Models: To further reduce costs and latency, experimenting with smaller models can be beneficial. Teammately Agents can attempt iterations with smaller architectures while maintaining the quality of output through continuous evaluations performed by LLM Judges. This approach can lead to a more efficient deployment that better aligns with resource constraints.
By considering these improvements, the AI's effectiveness, efficiency, and applicability across various scenarios and industries can be enhanced significantly.
Now it's your turn. Tell what AI you want to build.
AI Agent
Structured Output
Classification model
Marketing Engine