Pass JSON Schema directly to 'function' parameter

[benzourry]

Hi, is it possible to pass JSON Schema directly as function parameter without having to use @tool.
The use-case is that, I have various different JSON Schema stored in the database. So, I want to use that schema to produce different JSON response. JSON Schema can be updated and added by the user.

For now, I've just pass the schema as part of the user message. However, the result kind of hit or miss. I want to utilized the function call for data extraction with consistent JSON format structure.

https://nesin.io/blog/openai-api-json-response

I've seen example with data extraction with POJO, however I need to return a dynamically structured json using the JSON Schema that I already have.

[benzourry]

Update: I achieved it using model.generate() which accepts ToolSpecification

    public JsonNode extractToJson(String jsonSchemaString, String text) {
        ToolParameters params = ToolParameters.builder()
                .properties(objectMapper.readValue(jsonSchemaString, Map.class))
                .build();

        ToolSpecification spec = ToolSpecification.builder()
                .name("extract")
                .description("Extract to json")
                .parameters(params)
                .build();

        List<ChatMessage> messages = Collections.singletonList(
                new UserMessage("Extract json from text " + text)
        );

        return mapper.readTree(
                chatModel.generate(messages, spec).content().toolExecutionRequests().get(0).arguments()
        )
    }

I got it after sifting through issues and kind of got it from here: #601

However, it would be great if this can also be achieved with AiServices. I like the elegance of AiServices. :)

[langchain4j]

More info can be found here: https://docs.langchain4j.dev/tutorials/tools