langchaingo/examples/googleai-structured-output-example

Google AI Structured Output Example

This example shows how to get schema-constrained structured output from a Google Gemini model with the langchaingo library. Instead of parsing free-form text, you pass a JSON Schema and get back a JSON value that matches it.

What This Example Does

1. Sets up a Google AI (Gemini) client using your API key.
2. Defines a JSON Schema describing the shape it wants back (a Moon landing: mission, year, and an astronauts array).
3. Calls GenerateContent with llms.WithStructuredOutput(...), passing that schema.
4. Unmarshals the response straight into a typed Go struct and prints it.

How It Works

• llms.WithStructuredOutput(llms.StructuredOutputConfig{...}) is the provider-neutral way to request structured output. It carries a raw JSON Schema (Draft 2020-12); the SDK sends it to Gemini via responseJsonSchema and sets the response MIME type to application/json.
• On a normal completion the SDK validates the response against your original schema before returning it. So resp.Choices[0].Content is guaranteed to be a single JSON value matching the schema and unmarshals directly into a struct — no prompt engineering or manual cleanup.
• If the model returns something that does not match, you get a typed *llms.ErrStructuredOutputValidation instead of a silent bad result.

The same WithStructuredOutput option works across providers (OpenAI, Anthropic, Amazon Bedrock, Vertex, Ollama) — only the model and API key change.

Running the Example

1. Set your Google AI API key:

   export GOOGLE_API_KEY=your_api_key_here
   

2. Run the program:

   go run googleai-structured-output-example.go
   

3. Expected output (values come from the model):

   Mission:    Apollo 11
   Year:       1969
   Astronauts: [Neil Armstrong Buzz Aldrin Michael Collins]