Python's prompt word optimization solution based on DeepSeek big model

from typing import List, Dict

class DeepSeekPromptEngineer:
    """
     DeepSeek Structured Prompt Word Generation Tool
     Function: Build optimization prompt words that conform to MoE architecture based on input parameters
     """
    
    def __init__(self):
         = {
            "role": "Professional Role Definition",
            "task": "Core Task Description",
            "constraints": ["Time/Resource/Format Limit"],
            "steps": ["Step execution path"],
            "style": "Output Style Requirements"
        }

    def build_prompt(
        self,
        role: str,
        task: str,
        constraints: List[str],
        steps: List[str],
        style: str = "Professional and rigorous"
    ) -> Dict[str, str]:
        """
         Building structured prompt words
         :param role: role definition (such as 'Senior Data Analyst')
         :param task: task objective (need to include verb + object structure)
         :param constraints: 3D constraints (time/resource/quality)
         :param steps: step by step execution path
         :param style: output style
         """
        prompt = {
            "role": f"You are one{role}，The following tasks need to be completed：",
            "task": f"【Core mission】{task}",
            "constraints": "【Constraints】\n" + "\n".join([f"- {c}" for c in constraints]),
            "steps": "【Perform steps】\n" + "\n".join([f"{i+1}. {s}" for i, s in enumerate(steps)]),
            "style": f"【Output requirements】use{style}Style，useMarkdownFormat"
        }
        return "\n\n".join(())

#User Exampleif __name__ == "__main__":
    engineer = DeepSeekPromptEngineer()
    
    # Define financial analysis scenario parameters    example_prompt = engineer.build_prompt(
        role="Quantitative Financial Analyst",
        task="Analyze the investment risks of the new energy industry in Q2 2025",
        constraints=[
            "Data source: Wind Financial Terminal 2025Q1 data",
            "Time limit: Completed within 3 hours",
            "The output includes: policy/technology/market three dimensions"
        ],
        steps=[
            "Collect industry policy changes and subsidy data",
            "Calculate the correlation between the proportion of R&D investment of leading enterprises and PE",
            "Constructing Monte Carlo Simulated Risk Model",
            "Generate visual charts and executive summary"
        ],
        style="Brokerage research report style"
    )
    
    print("Created prompt word:\n")
    print(example_prompt)

# The template structure corresponds to the MoE architecture processing logic of DeepSeek = {
    "role": "Activate a network of experts in a specific field",  # Routing selection for MoE    "task": "Define attention focus goals",    # Control the allocation of information weights in the MLA mechanism    "constraints": "Set decision boundaries",   # Use three-dimensional constraints to reduce the model exploration space    "steps": "Activate reasoning ability in stages",   # Match the CoT characteristics of R1 model    "style": "Control Generate Distribution"          # Probability sampling strategy affecting the decoder}

def dynamic_adjustment(initial_output: str, feedback: str) -> str:
    """
     Implement dynamic optimization of prompt words
     :param initial_output: Initial generation result
     :param feedback: Fix the requirement (such as "The3Excessive step cost，Use machine learning solutions instead"）
    """
    adjustment_prompt = f"""
    Please optimize the solution based on the following feedback：
    [Original plan]
    {initial_output}
    
    [Revise requirements]
    {feedback}
    
    Require：
    1. 保留Original plan有效部分
    2. The version number must be marked in the modification part（v2.1）
    3. Explain the revised expected returns
    """
    return adjustment_prompt  # DeepSeek API should actually be called

def multimodal_prompt(image_desc: str, text_instruction: str) -> dict:
    """
     Construct multimodal prompt words
     :param image_desc: Image feature description (such as "CTImages show lower lobe of the right lung3cmNodal")
     :param text_instruction: text directive
     """
    return {
        "text": f"{text_instruction}\nBased on the following image features：{image_desc}",
        "visual_clues": [
            {"type": "medical_image", "region": "right_lower_lobe"},
            {"feature": "3cm_nodule", "confidence": 0.92}
        ],
        "constraints": [
            "Citizen "Lung Cancer Diagnosis and Treatment Guide 2025 Edition",
            "Requires TNM installment suggestions"
        ]
    }  # Activate DeepSeek's cross-modal alignment capability

role = "Blockchain Security Engineer"  # Trigger MoE routing in the field of network security

## Core Requirements- [x] Must include a zero-knowledge proof solution
- [ ] Exclude traditional encryption methods

steps = ["Proof of Concept → Prototyping → Stress Test"]  # Match the chain reasoning characteristics of R1

meta_prompt = "You are currently an expert in prompt word optimization, please upgrade the following instructions..."

"The generation plan must include: text report/3D model parameters/test data visualization"

"Highlights(Weight1.5)Cybersecurity Terms"