Launch a chat based recipe recommender that pulls the latest weekly menu from a public site, turns each recipe into vectors, and returns smart picks based on what users like and want to avoid. It is ideal for teams testing a product recommendation engine that improves selection and saves time.

The build path fetches the current week page with HTTP Request, then HTML and Code nodes parse the server data and collect the first set of courses. Extra HTML extraction grabs description, ingredients, utensils, and time. Set and Merge nodes assemble clean records with week and cuisine. Documents are split and embedded with Mistral AI, then stored in Qdrant. Full texts are also saved to a database for later retrieval. The chat path starts from a Chat Trigger, uses a Mistral chat model, and calls a tool that hits the Qdrant Recommend API with positive and negative signals. Grouping helps return unique recipes.

You will need a Mistral AI API key and a Qdrant collection sized 1024 with cosine distance. Point the nodes to your Qdrant endpoint, run the manual build to index the week, and open the chat URL to test. Expect faster menu curation, consistent answers, and a reusable pattern you can adapt to any catalog.

• Manual build grabs the current week page with HTTP Request.
• HTML node extracts server JSON and recipe details from the page.
• Code node parses data and selects the first ten courses.
• Set and Merge nodes create clean records with week, cuisine, and category.
• Recursive text splitter prepares long content for better embeddings.
• Mistral AI embeddings created with a native node and a direct API call, with a wait step for rate limits.
• Qdrant vector store saves vectors and uses the Recommend API with grouping to avoid duplicates.
• Chat Trigger and Mistral chat model power a live agent that calls a tool to fetch recommendations.
• Database steps save and fetch full recipes so the agent can return complete details.

• Reduce menu curation time from 2 hours to 5 minutes
• Automate about 90 percent of search and filtering work
• Improve match quality using positive and negative preferences
• Connect Mistral AI embeddings with Qdrant vector search
• Support hundreds of chat requests per day without extra staff

1. Import the template into n8n: Create a new workflow in n8n > Click the three dots menu > Select 'Import from File' > Choose the downloaded JSON file.
2. You'll need accounts with Mistral AI and Qdrant. See the Tools Required section above for links to create accounts with these services.
3. In your Mistral AI account, create an API key. In n8n, open Credentials > Create new credential > choose the Mistral Cloud API credential type > paste the API key and save. Assign this credential to the Embeddings Mistral Cloud, Mistral Cloud Chat Model, and Get Mistral Embeddings nodes.
4. In your Qdrant dashboard, create an API key and note your endpoint URL. In n8n, create a new Qdrant API credential with the endpoint and API key. Assign it to the Qdrant Vector Store and Use Qdrant Recommend API nodes.
5. In Qdrant, create a collection named hello_fresh with vector size 1024 and cosine distance. Confirm the collection exists before indexing.
6. Open the node Get This Week's Menu and confirm the URL uses the current year and week expression. Leave it as is if you want the latest week.
7. Run the workflow with the manual trigger to fetch the menu, parse recipes, embed texts, and store vectors in Qdrant. Check your Qdrant collection to confirm vectors were added.
8. Open the Chat Trigger node, copy the test URL, and send a message that lists what you want and what to avoid. Example: want spicy vegetarian dinner, avoid nuts.
9. Verify the agent calls the tool named Qdrant Recommend API. The tool embeds your positive and negative text, queries Qdrant, and returns grouped results. Check the Get Tool Response output for the raw payload.
10. If you get empty results, confirm the collection name matches, the embedding size is 1024, and your API keys are valid. Increase the Wait node value if you hit rate limits.
11. If parsing fails, review the HTML selectors in Extract Server Data and Extract Recipe Details. Site layout changes may require updating the CSS selectors.