This sample demonstrates an application that can understand a restaurant menu and answer relevant questions about the items on the menu.

There are 5 iterations of this sample application, growing in complexity and demonstrating utilization of many different Genkit features.

To test each one out, open the Developer UI and exercise the prompts and flows. Each step contains one or more example.json files which you can use as inputs.

Prerequisite

• Google Cloud project with Vertex AI API enabled (https://console.cloud.google.com/apis/library/aiplatform.googleapis.com)
• gcloud CLI installed (https://cloud.google.com/sdk/docs/install-sdk)
• to use Llama 3.1 405b enable it in the Vertex AI Model Garden

1. This step shows how to define prompts in code that can accept user input to their templates.
2. This step illustrates how to wrap your llm calls and other application code into flows with strong input and output schemas. It also adds an example of tool usage to load the menu from a data file.
3. This step adds session history and supports a multi-turn chat with the model.
4. This step ingests the menu items into a vector database and uses retrieval to include releveant menu items in the prompt.
5. This step illustrates how to combine models with different modalities. It uses a vision model to ingest the menu items from a photograph.

gcloud auth login
gcloud auth application-default login --project YOUR_PROJECT

npm i
npm run genkit:dev