atlas deployments setup

mongodb://localhost:12404/?directConnection=true

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

django-admin startproject cookbook --template https://github.com/mongodb-labs/django-mongodb-project/archive/refs/heads/5.0.x.zip
cd cookbook
python manage.py runserver

from django.db import models
from django_mongodb_backend.fields import ArrayField, EmbeddedModelField
from django_mongodb_backend.models import EmbeddedModel
from django_mongodb_backend.managers import MongoManager

class Features(EmbeddedModel):
    preparation_time = models.CharField(max_length=100)
    complexity = models.CharField(max_length=100)
    prep_time = models.IntegerField()
    cuisine = models.CharField(max_length=100, null=True, blank=True)

class Recipe(models.Model):
    title = models.CharField(max_length=200)
    instructions = models.TextField(blank=True)
    features = EmbeddedModelField(Features, null=True, blank=True)
    ingredients = ArrayField(models.CharField(max_length=100), null=True, blank=True)
    embedding_ingredients = models.CharField(max_length=500, null=True, blank=True)
    voyage_embedding = models.JSONField(null=True, blank=True)
    objects = MongoManager()

    class Meta:
        db_table = "recipes"
        managed = False

    def __str__(self):
        return f"Recipe {self.title}"

def recipe_statistics(request):
    pipeline = [
        {"$project": {"_id": 1, "cuisine": "$features.cuisine"}},
        {"$group": {"_id": "$cuisine", "count": {"$sum": 1}}},
        {"$sort": {"count": -1}},
        {
            "$project": {
                "_id": 1,
                "cuisine": {"$ifNull": ["$_id", "Unspecified"]},
                "count": 1,
            }
        }
    ]
    stats = Recipe.objects.raw_aggregate(pipeline)
    result = list(stats)
    return render(request, "statistics.html", {"cuisine_stats": result})

{
  "fields": [
    {
      "type": "vector",
      "path": "voyage_embedding",
      "numDimensions": 1024,
      "similarity": "cosine"
    }
  ]
}

def perform_vector_search(query_text, limit=10, num_candidates=None):
    # Khởi tạo VoyagerAI Client, tạo embedding cho query
    # Thực thi raw_aggregate với $vectorSearch pipeline trên MongoDB
    # Trả về danh sách recipe có điểm tương đồng cao nhất

def get_claude_suggestions(user_ingredients, similar_recipes, max_suggestions=4):
    # Tạo prompt chi tiết
    # Gọi Anthropic Claude API với prompt
    # Xử lý kết quả trả về thành danh sách đề xuất