mscheong01 / interfaice Goto Github PK

ex)

interface TestInterface {
    @OpenAiChat
    fun randomCityNameInCountryOrNull(cityName: String?): String?
}

I think we missed out Array types when adding support for collection types (currently, only List, Set and Map is supported)

I think some users want to enrich auto generated prompt for user-defined classes.

Maybe we can add an annotation for it. I will work on it right away after the PR about ObjectRule is merged.

Because the gradle script of spring-boot-starter-interfAIce doesn't include from(components["java"]), it doesn't generate jar file. And I think it's better to include sources and javadocs for the module.

Also, I think we should rename the module to interfAIce-spring-boot-starter, because the official doc of Spring Boot says 'Do not start your module names with spring-boot, even if you use a different Maven groupId.'.

ex)

interface TestInterface {
    @OpenAiChat
    fun getGreetingCard(person: Person): GreetingCard
}

class Person(
    val name: String,
    val age: Int,
    val locale: String,
)

class GreetingCard(
    val message: String,
)

When executing a Test where an error occurs in InvocationHandler, it takes a while for the system to exit after the exception is thrown. It looks like the probable cause is that the background thread that calls the openai api is not cleaned up and runs regardless of the error. This did not happen when I used OpenAiApiClientdefined in the spring boot starter that uses a spring webclient.
suspicious code:
https://github.com/mscheong01/interfAIce/blob/main/core/src/main/kotlin/io/github/mscheong01/interfaice/openai/DefaultOkHttpOpenAiClient.kt

Mono.create { sink ->
    client.newCall(httpRequest).enqueue(object : com.squareup.okhttp.Callback {
        override fun onFailure(request: Request, e: java.io.IOException) {
            sink.error(e)
        }

        override fun onResponse(response: com.squareup.okhttp.Response) {
            try {
                if (!response.isSuccessful) {
                    throw IOException("Error: ${response.code()} ${response.message()}")
                } else {
                    sink.success(mapper.readValue(response.body().string()))
                }
            } catch (e: Exception) {
                sink.error(e)
            }
        }
    })
}

OpenAI has added "function calling" feature to the Chat api
https://openai.com/blog/function-calling-and-other-api-updates

We could leverage this feature to allow interfAIce proxies to call upon other methods
ex)

interface FunctionAware<T> {
    fun withFunction( ... ): T
    fun withFunctions( ... ): T
}
interface WeatherService : FunctionAware<WeatherService> {
    fun getWeather(): Weather
}

val proxy = OpenAiProxyFactory.of("sk-****").create<WeatherService>()
proxy
    .withFunction(weatherClient::greet)
    .getWeather()

I think we can utilize internal caches for not executing reflection and many logic every time. This is surely optional feature, I think we can add it after the initial release. I'm interested in the implementation.