from

langchain_core.prompts import PromptTemplate

template = PromptTemplate.from_template("You are a helpful assistant. Explain {topic} to a 5-year-old.")

We can reuse this for 'Space', 'Economics', or 'Cooking'!

**2. ChatPromptTemplate (The Gold Standard)**  
Modern chat models require explicit role segregation. This template structures messages into system, human, and assistant roles, ensuring the model correctly interprets instructions, user input, and behavioral constraints.

from langchain_core.prompts import ChatPromptTemplate

chat_template = ChatPromptTemplate.from_messages([ ("system", "You are a professional {industry} consultant."), ("human", "How can I improve my {business_process}?"), ])

When we run this, LangChain formats it perfectly for the AI

formatted_prompt = chat_template.invoke({ "industry": "Real Estate", "business_process": "lead generation" })

**Role Architecture:**
- **System**: Defines behavioral constraints, tone, output format, and safety boundaries.
- **Human**: Captures dynamic user input or application-generated queries.
- **AI/Assistant**: Enables few-shot prompting by providing example responses, drastically improving output consistency.

**Partial Formatting Strategy:**  
When certain variables are known at initialization time but others arrive at runtime, partial formatting decouples template construction from invocation. This pattern reduces redundant template recompilation and enforces modular chain design.

Create a template with a fixed 'name' but a dynamic 'question'

partial_template = chat_template.partial(industry="Tech Operations")

Now you only need to provide the 'business_process' later!

final_chain = partial_template | model

## Pitfall Guide
1. **Mixing Completion and Chat Paradigms**: Using `PromptTemplate` with chat-optimized models strips role metadata, causing the model to treat system instructions as generic user text. Always match template type to model architecture.
2. **Over-Partialing Without Runtime Validation**: Partially formatting templates can mask missing variables until invocation. Implement explicit validation or use `invoke()` with strict key checking to prevent silent `KeyError` failures in production.
3. **Ignoring Role Boundary Enforcement**: Failing to separate system constraints from human input leads to prompt injection and instruction bleeding. Always isolate behavioral rules in the `system` role and user data in the `human` role.
4. **Hardcoding Dynamic Context**: Embedding dates, session IDs, or user metadata directly into templates breaks reusability and inflates token usage. Extract all variable data into placeholders and inject at runtime.
5. **Token Budget Mismanagement**: Unstructured templates often repeat formatting instructions or include unnecessary conversational filler. Audit templates for token efficiency; use concise system prompts and leverage few-shot examples only when output consistency is critical.
6. **Skipping Few-Shot Alignment**: Omitting the `AI/Assistant` role when complex formatting is required forces the model to guess output structure. Provide 2-3 high-quality examples in the assistant role to lock in JSON, markdown, or list formatting.

## Deliverables
- **📘 Prompt Template Architecture Blueprint**: A decision matrix for selecting between `PromptTemplate` and `ChatPromptTemplate`, role mapping guidelines, and partial formatting chain diagrams for scalable AI pipelines.
- **✅ Pre-Deployment Prompt Validation Checklist**: 12-point audit covering variable injection safety, role boundary integrity, token budget limits, injection resistance, and fallback behavior for missing parameters.
- **⚙️ Configuration Templates**: Production-ready `ChatPromptTemplate` configurations for common use cases (Code Reviewer, Data Analyst, Customer Support), including system role constraints, human input slots, and few-shot assistant examples.