Building AI-Powered Applications with Generative Model APIs

Choosing Your API

The main choices for production applications: OpenAI (best ecosystem, easy start), Anthropic (best for complex instructions and safety), Google Gemini (best for multimodal and long context), and open-source models via Ollama/Together.ai (best for cost and privacy).

Text Generation with Streaming

import openai

client = openai.OpenAI()

# Streaming — display tokens as they arrive
with client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Write a haiku about AI agents."}],
    stream=True
) as stream:
    for chunk in stream:
        if chunk.choices[0].delta.content:
            print(chunk.choices[0].delta.content, end="", flush=True)

Image Generation

# DALL-E 3
response = client.images.generate(
    model="dall-e-3",
    prompt="A photorealistic image of a robot reading a book in a cosy library",
    size="1792x1024",
    quality="hd",
    n=1
)
image_url = response.data[0].url

# Stability AI (local-compatible models)
import requests
resp = requests.post(
    "https://api.stability.ai/v1/generation/stable-diffusion-xl-1024-v1-0/text-to-image",
    headers={"Authorization": f"Bearer {api_key}"},
    json={"text_prompts": [{"text": prompt}], "width": 1024, "height": 1024}
)

Robust Error Handling

import openai
from tenacity import retry, stop_after_attempt, wait_exponential

@retry(stop=stop_after_attempt(3), wait=wait_exponential(min=1, max=10))
def generate_text(prompt: str) -> str:
    try:
        resp = client.chat.completions.create(model="gpt-4o-mini",
            messages=[{"role":"user","content":prompt}])
        return resp.choices[0].message.content
    except openai.RateLimitError:
        raise  # Let tenacity handle it
    except openai.APIError as e:
        return f"Generation failed: {str(e)}"

Always implement exponential backoff and fallbacks. At scale, even a 99.9% uptime API will fail you multiple times per day.

Cost Optimisation at Scale

• Cache frequent responses — same prompt, same output, no need to call the API twice.
• Use smaller models for classification and routing, larger ones for generation.
• Compress prompts — reduce system prompt length by 30% with no quality loss using prompt compression.
• Batch requests where latency is not critical — batch APIs are typically 50% cheaper.