One Language forAI-Native Full-StackDevelopment

import json, base64
from datetime import datetime
from openai import OpenAI

client = OpenAI()

# --- Lots of boilerplate just to define a schema ---
tools = [{
    "type": "function",
    "function": {
        "name": "process_memory",
        "description": 
            "Extract structured memory details from the photo",
        "parameters": {
            "type": "object",
            "properties": {
                "who": {
                    "type": "array",
                    "items": {"type": "string"},
                    "description": "Names of people in the photo"
                },
                "what": {
                    "type": "string",
                    "description": "What is happening in the scene"
                },
                "where": {
                    "type": "string",
                    "description": 
                        "Location or setting of the photo"
                }
            },
            "required": ["who", "what", "where"]
        }
    }
}]

# --- The prompt has to sit here like a Novel ---
SYS_PROMPT = """
# Role and Objective
Your goal is to extract structured memory details from
referenced images and user context. Engage in a way that feels 
natural, as if you were helping someone document their 
experiences, but always return structured output.

# Instructions
- Extract details only based on the image and user input.
- Avoid hallucinations or assumptions if information is missing.
- Always call the `process_memory` tool to return results.
- Keep results short, factual, and consistent.

# Sub-categories for more detailed instructions
## First Turn
- React to the image or context briefly.
- Encourage clarification if information is incomplete.

## Field Writing (for process_memory)
- **who**: list of people, animals, or notable entities.
- **what**: concise description of the activity or event 
  (3-5 words).
- **where**: specific place, city, or landmark provided 
   or visible.

### Reasoning Steps
- If the user corrects existing information, update the fields.
- If new information is provided, add it without discarding 
  existing relevant details.
- If a field is not mentioned or visible, leave it empty.

## Response Writing
- Keep responses factual and grounded in what's visible or given.
- Never ask the user irrelevant questions.
- Use the photo context to refine details when applicable.

# Output Format
Always return JSON via the `process_memory` tool with:
- `who`: list of strings
- `what`: string (short activity/event description)
- `where`: string (location or place)
"""

with open("image.png", "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode("utf-8")

# --- Verbose message construction ---
messages = [
    {"role": "system", "content": SYS_PROMPT},
    {
        "role": "user",
        "content": [
            {"type": "text", "text": "Photo took in Paris."},
            {"type": "image_url", "image_url": {
                "url": f"data:image/png;base64,{image_b64}"}
            }
        ]
    }
]

response = client.chat.completions.create(
    model="gpt-4o",
    messages=messages,
    tools=tools,
    tool_choice={
        "type": "function", 
        "function": {"name": "process_memory"}
    }
)

result = json.loads(
    response.choices[0].message.tool_calls[0].function.arguments
)

print(result)