18. Reading API Responses

JSON Response Structure

AI APIs return structured JSON. Extract data carefully:

response = {
    "id": "chatcmpl-123",
    "model": "gpt-4",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "Hello! How can I help you?"
            },
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 10,
        "completion_tokens": 12,
        "total_tokens": 22
    }
}

# Extract nested values
content = response["choices"][0]["message"]["content"]
usage = response["usage"]["total_tokens"]

Handling Missing Keys

# Safe extraction with defaults
def get_content(response):
    try:
        return response.get("choices", [{}])[0].get("message", {}).get("content", "")
    except (KeyError, IndexError, TypeError):
        return ""

# Or use .get() chaining
content = (
    response
    .get("choices", [{}])[0]
    .get("message", {})
    .get("content", "")
)

Batch Responses

AI APIs return lists of completions:

responses = [
    {"choices": [{"message": {"content": "Response 1"}}]},
    {"choices": [{"message": {"content": "Response 2"}}]},
    {"choices": [{"message": {"content": "Response 3"}}]}
]

contents = [r["choices"][0]["message"]["content"] for r in responses]

Pagination

Large results come in pages:

all_results = []
page_token = None

while True:
    params = {"limit": 100}
    if page_token:
        params["after"] = page_token
    
    response = requests.get(url, headers=headers, params=params).json()
    all_results.extend(response.get("data", []))
    
    page_token = response.get("next_cursor")
    if not page_token:
        break

Local verification checkpoint

Run the smallest example from this chapter in a local workspace and record the package version, runtime, data path, and observed output. If the result depends on model size, vector count, CPU/GPU backend, or available memory, note that constraint beside the exercise so the lesson remains reproducible.

Local verification checkpoint

Run the smallest example from this chapter in a local workspace and record the package version, runtime, data path, and observed output. If the result depends on model size, vector count, CPU/GPU backend, or available memory, note that constraint beside the exercise so the lesson remains reproducible.