prompt-templates

建設 AI 向けプロンプトテンプレート

概要

建設業界の AI タスクに最適化された、構造化された再利用可能なプロンプトテンプレートです。これらのテンプレートにより、コスト見積もり、工程分析、文書処理、BIM データへの問い合わせで、一貫性のある高品質な出力を実現できます。

テンプレートのフレームワーク

基本テンプレートの構造

from dataclasses import dataclass, field
from typing import Dict, Any, List, Optional
from string import Template
import json

@dataclass
class PromptTemplate:
    name: str
    description: str
    template: str
    input_variables: List[str]
    output_format: Optional[str] = None
    examples: List[Dict[str, Any]] = field(default_factory=list)
    category: str = "general"
    version: str = "1.0"

    def format(self, **kwargs) -> str:
        """Format template with provided variables."""
        # Validate all required variables are provided
        missing = [v for v in self.input_variables if v not in kwargs]
        if missing:
            raise ValueError(f"Missing required variables: {missing}")

        # Format template
        prompt = Template(self.template).safe_substitute(**kwargs)

        # Add output format if specified
        if self.output_format:
            prompt += f"\n\nOutput Format:\n{self.output_format}"

        return prompt

    def with_examples(self, n: int = 2) -> str:
        """Return template with few-shot examples."""
        examples_text = ""
        for i, ex in enumerate(self.examples[:n], 1):
            examples_text += f"\nExample {i}:\n"
            examples_text += f"Input: {ex.get('input', '')}\n"
            examples_text += f"Output: {ex.get('output', '')}\n"

        return f"{examples_text}\n{self.template}"


class ConstructionPromptLibrary:
    """Library of construction-specific prompt templates."""

    def __init__(self):
        self.templates: Dict[str, PromptTemplate] = {}
        self._register_defaults()

    def register(self, template: PromptTemplate):
        self.templates[template.name] = template

    def get(self, name: str) -> Optional[PromptTemplate]:
        return self.templates.get(name)

    def list_by_category(self, category: str) -> List[PromptTemplate]:
        return [t for t in self.templates.values() if t.category == category]

    def _register_defaults(self):
        # Register all default templates
        for template in DEFAULT_TEMPLATES:
            self.register(template)

コスト見積もりテンプレート

項目分類

COST_LINE_ITEM_CLASSIFICATION = PromptTemplate(
    name="cost_line_item_classification",
    description="Classify cost estimate line items to CSI MasterFormat divisions",
    category="cost_estimation",
    input_variables=["line_items"],
    template="""You are a construction cost estimator. Classify each line item to its appropriate CSI MasterFormat division.

Line Items to Classify:
$line_items

For each line item, provide:
1. CSI Division (2-digit)
2. CSI Section (6-digit code)
3. Confidence level (high/medium/low)

Use standard CSI MasterFormat 2020 divisions:
- 03: Concrete
- 04: Masonry
- 05: Metals
- 06: Wood, Plastics, Composites
- 07: Thermal and Moisture Protection
- 08: Openings
- 09: Finishes
- 22: Plumbing
- 23: HVAC
- 26: Electrical
- 31: Earthwork
- 32: Exterior Improvements
- 33: Utilities
""",
    output_format="""JSON array with format:
[
  {
    "line_item": "original description",
    "csi_division": "XX",
    "csi_section": "XX XX XX",
    "csi_title": "Section Title",
    "confidence": "high|medium|low"
  }
]""",
    examples=[
        {
            "input": "4000 PSI concrete for foundations",
            "output": '{"csi_division": "03", "csi_section": "03 30 00", "csi_title": "Cast-in-Place Concrete", "confidence": "high"}'
        }
    ]
)

単価の検証

COST_UNIT_PRICE_VALIDATION = PromptTemplate(
    name="cost_unit_price_validation",
    description="Validate unit prices against industry standards",
    category="cost_estimation",
    input_variables=["line_items", "location", "year"],
    template="""You are a construction cost analyst. Review these unit prices for reasonableness.

Project Location: $location
Cost Year: $year

Line Items to Review:
$line_items

For each line item:
1. Assess if the unit price is reasonable for the location and year
2. Flag any outliers (too high or too low)
3. Suggest corrections if needed

Consider regional cost factors, labor rates, and material costs for $location.
""",
    output_format="""JSON array:
[
  {
    "line_item": "description",
    "current_unit_cost": 0.00,
    "assessment": "reasonable|high|low",
    "typical_range": {"low": 0.00, "high": 0.00},
    "suggested_unit_cost": 0.00,
    "notes": "explanation"
  }
]"""
)

見積もり概要の生成

COST_ESTIMATE_SUMMARY = PromptTemplate(
    name="cost_estimate_summary",
    description="Generate executive summary from detailed estimate",
    category="cost_estimation",
    input_variables=["project_name", "estimate_data", "gross_area"],
    template="""Generate an executive summary for this construction cost estimate.

Project: $project_name
Gross Area: $gross_area SF

Estimate Data:
$estimate_data

Create a professional summary including:
1. Total project cost and cost per SF
2. Major cost drivers (top 5 divisions)
3. Key assumptions and exclusions
4. Risk factors affecting the estimate
5. Recommendations for cost optimization

Write in a professional tone suitable for owner/stakeholder presentation.
""",
    output_format="""Markdown format with sections:
## Executive Summary
## Cost Breakdown
## Key Assumptions
## Risk Factors
## Recommendations"""
)

工程分析テンプレート

クリティカルパス分析

SCHEDULE_CRITICAL_PATH = PromptTemplate(
    name="schedule_critical_path_analysis",
    description="Analyze schedule critical path and float",
    category="scheduling",
    input_variables=["schedule_data", "data_date"],
    template="""Analyze the critical path for this construction schedule.

Data Date: $data_date

Schedule Data:
$schedule_data

Provide analysis of:
1. Current critical path activities
2. Near-critical activities (total float < 10 days)
3. Float consumption trends
4. Schedule risk areas
5. Recommendations for protecting the critical path

Focus on activities that could impact the project completion date.
""",
    output_format="""JSON:
{
  "critical_path": [
    {"activity_id": "", "name": "", "duration": 0, "total_float": 0}
  ],
  "near_critical": [...],
  "risk_areas": ["description"],
  "recommendations": ["action item"]
}"""
)

遅延分析

SCHEDULE_DELAY_ANALYSIS = PromptTemplate(
    name="schedule_delay_analysis",
    description="Analyze schedule delays and impacts",
    category="scheduling",
    input_variables=["baseline_schedule", "current_schedule", "delay_events"],
    template="""Perform a delay analysis comparing baseline to current schedule.

Baseline Schedule:
$baseline_schedule

Current Schedule:
$current_schedule

Known Delay Events:
$delay_events

Analyze:
1. Total project delay in calendar days
2. Critical path delays vs non-critical delays
3. Concurrent delays
4. Pacing delays
5. Attribution of delays (owner, contractor, third-party, weather)

Use the Time Impact Analysis (TIA) methodology.
""",
    output_format="""JSON:
{
  "total_delay_days": 0,
  "delay_breakdown": {
    "owner_caused": 0,
    "contractor_caused": 0,
    "concurrent": 0,
    "excusable": 0
  },
  "impacted_milestones": [...],
  "recommendations": [...]
}"""
)

リソース平準化

SCHEDULE_RESOURCE_LEVELING = PromptTemplate(
    name="schedule_resource_leveling",
    description="Suggest resource leveling for over-allocated resources",
    category="scheduling",
    input_variables=["schedule_data", "resource_limits"],
    template="""Analyze resource allocation and suggest leveling strategies.

Schedule Data:
$schedule_data

Resource Limits:
$resource_limits

Identify:
1. Over-allocated resources by period
2. Peak demand periods
3. Activities that can be shifted (have float)
4. Recommended activity adjustments

Maintain critical path while optimizing resource usage.
""",
    output_format="""JSON:
{
  "overallocations": [
    {"resource": "", "period": "", "demand": 0, "limit": 0}
  ],
  "recommendations": [
    {"activity_id": "", "action": "shift|split|extend", "days": 0, "reason": ""}
  ],
  "revised_peak_demand": {}
}"""
)

文書処理テンプレート

RFI 回答の生成

DOC_RFI_RESPONSE = PromptTemplate(
    name="doc_rfi_response",
    description="Generate professional RFI response",
    category="document_processing",
    input_variables=["rfi_number", "question", "context", "spec_references"],
    template="""Generate a professional response to this RFI.

RFI Number: $rfi_number

Question:
$question

Context/Background:
$context

Relevant Specification References:
$spec_references

Write a clear, professional response that:
1. Directly addresses the question
2. References relevant specifications
3. Provides clear direction
4. Notes any cost or schedule implications
5. Identifies if further clarification is needed

Maintain a professional, unambiguous tone.
""",
    output_format="""Response in professional letter format with:
- Direct answer to the question
- Supporting references
- Required actions (if any)
- Disclaimer about cost/schedule (if applicable)"""
)

仕様の抽出

DOC_SPEC_EXTRACTION = PromptTemplate(
    name="doc_spec_extraction",
    description="Extract key requirements from specifications",
    category="document_processing",
    input_variables=["spec_text", "section_number"],
    template="""Extract key requirements from this specification section.

Section Number: $section_number

Specification Text:
$spec_text

Extract and categorize:
1. Material requirements (products, standards, grades)
2. Performance requirements (strength, ratings, certifications)
3. Submittal requirements (shop drawings, samples, certifications)
4. Installation requirements (methods, tolerances, conditions)
5. Quality control requirements (testing, inspections)
6. Warranty requirements

Flag any ambiguous or conflicting requirements.
""",
    output_format="""JSON:
{
  "section": "XX XX XX",
  "title": "",
  "materials": [...],
  "performance": [...],
  "submittals": [...],
  "installation": [...],
  "quality_control": [...],
  "warranty": {...},
  "ambiguities": [...]
}"""
)

変更指示の正当性説明

DOC_CHANGE_ORDER = PromptTemplate(
    name="doc_change_order_justification",
    description="Generate change order justification narrative",
    category="document_processing",
    input_variables=["change_description", "cause", "cost_impact", "schedule_impact", "supporting_docs"],
    template="""Generate a professional change order justification.

Change Description:
$change_description

Cause of Change:
$cause

Cost Impact: $cost_impact
Schedule Impact: $schedule_impact

Supporting Documentation:
$supporting_docs

Write a clear justification that:
1. Describes the change and its necessity
2. Explains why this was not included in the original scope
3. Details the cost basis (labor, material, equipment, markup)
4. Explains schedule impact and mitigation
5. References supporting documentation

Maintain a factual, professional tone suitable for owner review.
"""
)

BIM 問い合わせテンプレート

要素数量の抽出

BIM_QTO_EXTRACTION = PromptTemplate(
    name="bim_qto_extraction",
    description="Extract quantities from BIM element data",
    category="bim",
    input_variables=["element_data", "quantity_types"],
    template="""Extract quantities from this BIM element data.

Element Data:
$element_data

Quantity Types Needed:
$quantity_types

For each element:
1. Identify the element type and classification
2. Extract requested quantities
3. Apply appropriate unit conversions
4. Group by classification code

Ensure quantities are in standard construction units (SF, LF, CY, EA, etc.).
""",
    output_format="""JSON:
{
  "summary": [
    {
      "classification": "XX XX XX",
      "description": "",
      "quantity": 0.00,
      "unit": "",
      "element_count": 0
    }
  ],
  "details": [...]
}"""
)

干渉分析の概要

BIM_CLASH_ANALYSIS = PromptTemplate(
    name="bim_clash_analysis",
    description="Summarize and prioritize BIM clash detection results",
    category="bim",
    input_variables=["clash_data", "disciplines"],
    template="""Analyze and summarize these BIM clash detection results.

Clash Data:
$clash_data

Disciplines Involved: $disciplines

Provide:
1. Summary of clash counts by discipline pair
2. Priority ranking of clashes (critical/major/minor)
3. Recommended resolution responsibility
4. Pattern analysis (recurring issues)
5. Suggested coordination meetings

Focus on actionable insights for the coordination team.
""",
    output_format="""JSON:
{
  "total_clashes": 0,
  "by_discipline_pair": {...},
  "by_priority": {"critical": 0, "major": 0, "minor": 0},
  "patterns": [...],
  "recommendations": [...]
}"""
)

テンプレートの使用方法

# Initialize library
library = ConstructionPromptLibrary()

# Get template
template = library.get("cost_line_item_classification")

# Format with variables
prompt = template.format(
    line_items="""
    1. 4000 PSI structural concrete
    2. #5 rebar
    3. CMU block wall
    4. Steel wide flange beams
    5. Spray foam insulation
    """
)

# Use with LLM
response = llm.generate(prompt)

# With few-shot examples
prompt_with_examples = template.with_examples(n=2)

カスタムテンプレートの作成

# Create project-specific template
my_template = PromptTemplate(
    name="my_project_template",
    description="Custom template for specific project needs",
    category="custom",
    input_variables=["var1", "var2"],
    template="""
    Your custom prompt here with $var1 and $var2 placeholders.
    """,
    output_format="JSON or other format specification"
)

# Register to library
library.register(my_template)

ベストプラクティス

1. 具体的にする：分野のコンテキスト（CSI コード、建設用語）を含めます。
2. 出力形式を定義する：想定される JSON／Markdown の構造を必ず指定します。
3. 例を含める：少数ショットの例により、一貫性が向上します。
4. テンプレートをバージョン管理する：再現性を確保するために変更を追跡します。
5. 十分にテストする：出力が想定される形式と一致することを検証します。

リソース

• プロンプトエンジニアリングガイド：https://www.promptingguide.ai/
• CSI MasterFormat：https://www.csiresources.org/standards/masterformat
• 建設用語集：AIA、AGC、CMAA の用語集