Skip to content

chain7.md

Latest commit

 

History

History
86 lines (73 loc) · 3.67 KB

chain7.md

File metadata and controls

86 lines (73 loc) · 3.67 KB
Raw

Chain7 Developer Documentation

Overview

" Advanced prompt composition using PipelinePromptTemplate for complex multi-part prompts " Movie recommendation AI system with structured example-based responses " Modular prompt architecture with reusable components

Key Features

" Pipeline Prompts: Composition of multiple PromptTemplate instances " Structured AI Responses: Example-driven conversation formatting " Modular Design: Separate prompts for system role, examples, and user input " Variable Interpolation: Multiple template variables across pipeline stages

Architecture Components

" Model: Google Gemini 1.5 Flash for movie recommendations " System Prompt: AI assistant role definition as movie buff " Example Response: Template showing desired response format " Conversation Prompt: User query template for movie recommendations " Final Prompt: Composition template combining all components

Pipeline Structure

" systemRole: "You are a helpful A.I assistant who is also a movie buff" " aiExampleResponse: Format template with example genre and answers " newConversation: User question template for specific genre requests " finalHumanPrompt: Combines all components with placeholder variables

Prompt Composition Pattern

const composedPrompt = new PipelinePromptTemplate({
    finalPrompt: finalHumanPrompt,
    pipelinePrompts: [
        { name: "systemRole", prompt: systemPrompt },
        { name: "aiExampleResponse", prompt: aiExampleResponsePrompt },
        { name: "newConversation", prompt: newConversationPrompt }
    ]
});

Variable System

" example_genre: Genre used in example formatting (e.g., "action") " example_answer: Movie title for example responses (e.g., "Mission Impossible") " question_genre: Target genre for user recommendation request " Template Inheritance: Variables flow through pipeline components

Execution Methods

" Direct Format: composedPrompt.format() then model.invoke() " Chain Pipeline: composedPrompt.pipe(model).invoke() (recommended) " Commented Alternative: Shows both approaches for comparison

Response Format Structure

" Question Format: "Suggest 3 most recent {genre} movie?" " Answer Pattern: "Great question, {movie} {movie} {movie} is my favorite" " Structured Output: Consistent formatting across all movie recommendations

Configuration

" Environment: Requires GOOGLE_API_KEY for Google GenAI access " Model: Gemini 1.5 Flash for conversational responses " Temperature: Default setting (not explicitly set)

Developer Notes

" Prompt Hierarchy: Final prompt combines all pipeline components " Variable Scoping: Each pipeline prompt can have independent variables " Format Consistency: Example response guides AI output structure " Chain vs Direct: Pipeline method preferred for complex prompt composition " Reusability: Individual prompt templates can be reused in other contexts

Use Cases

" Complex Prompt Engineering: Multi-component prompt systems " Conversational AI: Structured response formatting " Domain-Specific Assistants: Movie recommendation specialization " Template Composition: Building prompts from modular components

Example Usage

const response = await chain.invoke({
    example_genre: "action",
    example_answer: "Mission Impossible", 
    question_genre: "science fiction"
});

Benefits

" Maintainability: Separate concerns for different prompt components " Flexibility: Easy modification of individual prompt parts " Consistency: Structured approach to complex prompt creation " Reusability: Modular components for different conversation types