Generating Website Content with Amazon Nova Lite

The Text Model

Amazon Nova Lite (amazon.nova-lite-v1:0) runs on Bedrock in us-east-1. It's fast (~15 seconds for a full site) and inexpensive. The payload format follows the Bedrock Converse API style with a messages array.

TEXT_MODEL = "amazon.nova-lite-v1:0"

def _generate_text(prompt: str) -> str:
    client = boto3.client(
        "bedrock-runtime",
        region_name="us-east-1",
        aws_access_key_id=config("AWS_ACCESS_KEY_ID"),
        aws_secret_access_key=config("AWS_SECRET_ACCESS_KEY"),
    )
    response = client.invoke_model(
        modelId=TEXT_MODEL,
        contentType="application/json",
        accept="application/json",
        body=json.dumps({
            "messages": [{"role": "user", "content": [{"text": prompt}]}],
            "inferenceConfig": {"max_new_tokens": 2000, "temperature": 0.8},
        }),
    )
    body = json.loads(response["body"].read())
    return body["output"]["message"]["content"][0]["text"]

The Prompt Structure

The prompt explicitly requests a raw JSON object with no markdown wrapping. Nova Lite sometimes wraps JSON in triple-backtick blocks anyway — the parser strips those before calling json.loads.

def generate_text_only(site) -> dict:
    raw = _generate_text(prompt).strip()
    if raw.startswith("```"):
        raw = raw.split("```")[1]
        if raw.startswith("json"):
            raw = raw[4:]
    content = json.loads(raw.strip())

Handling Array Content

Nova Lite sometimes returns blog post content as an array of paragraph strings instead of a single string. The normalisation step joins them before storing.

for blog_type in content.get("blog_types", []):
    for post in blog_type.get("posts", []):
        if isinstance(post.get("content"), list):
            post["content"] = "\n\n".join(post["content"])