openai-gpt4

OpenAI GPT-4 Chat Completion Pipeline

You need GPT-4 in a production workflow, but prompt construction, HTTP calls, and response parsing are tangled in one method. This three-step pipeline separates request building (with system message support), API invocation (via a shared OpenAiService), and result extraction.

Workflow

prompt, systemMessage, model
       │
       ▼
┌──────────────────────┐
│ gpt4_build_request   │  Build messages array with system + user
└──────────┬───────────┘
           │  requestBody {model, messages, max_tokens, temperature}
           ▼
┌──────────────────────┐
│ gpt4_call_api        │  Call OpenAI Chat Completions
└──────────┬───────────┘
           │  apiResponse {choices, usage}
           ▼
┌──────────────────────┐
│ gpt4_extract_result  │  Extract text from choices[0]
└──────────────────────┘
           │
           ▼
     result, model, usage

Workers

Gpt4BuildRequestWorker (gpt4_build_request) -- Constructs a messages array with a system message and user message as List.of(systemMsg, userMsg). Defaults model to gpt-4o-mini via configuredDefaultModel() which checks the OPENAI_MODEL env var. Coerces max_tokens and temperature from workflow input.

Gpt4CallApiWorker (gpt4_call_api) -- Uses a shared OpenAiService class for the HTTP call. Extracts the model name from requestBody.getOrDefault("model", DEFAULT_MODEL). Constructs the apiResponse with a choices list containing the response.

Gpt4ExtractResultWorker (gpt4_extract_result) -- Pulls the generated text from the API response structure and returns it as the final result.

Tests

13 tests cover request building with system messages, API calling, and result extraction.

Further Reading

• RUNNING.md -- how to build and run this example