PromptTemplateManager

Purpose

Context in Monorepo

The PromptTemplateManager will be the second available module in the set and sequence of Data Augmentation Workflow with Microfrontends.
The goal of the Augment-It Monorepo is to transform and augment a set of records by using various AI services. For instance, AI-Powered Search models, AI Powered Data Capture techniques, such as AI Web Crawlers and Web Scrapers through their respective APIs.
By properly integrating the PromptTemplateManager with the RecordCollector, we better ensure prompts well leverage AI capabilities and augment specific customer data effectively.
graph TD subgraph "PromptTemplateManager" A[UI Layer] -->|Manages| B[Template Editor] A -->|Manages| C[Variable Mapper] A -->|Manages| D[Template Library] B -->|Uses| E[MDX Editor] C -->|Pulls from| F[RecordCollector API] D -->|Stores| G[Template Storage] end subgraph "External Systems" F -->|Provides| H[Customer Data] I[AI Services] -->|Receives| J[Populated Templates] end C -->|Maps variables to| B B -->|Saves to| D D -->|Loads to| B D -->|Sends populated templates to| I F -->|Syncs with| H

Functionality of PromptTemplateManager

The PromptTemplateManager will provide functionalities to allow users to upload prompts, author new ones, and insert managed variables called from the records made available from the RecordCollector.
ℹ️
The initial use case is using the RecordCollector to pull records from the CRM system about customers.
This data may be mediated through DataBricks (or another data aggregator service).
The PromptTemplateManager shall be the tool that empowers users to select, generate, and iterate on meaningful prompts likely to have the best results. These prompts will specifically make easy the addition of fields as variables to allow the generation of customer specific prompts from a template.
The PromptTemplateManager will enable the users to select the most appropriate prompt to send to the RequestReviewer, which may or may not be the one they were just creating, working on, or saved to disk.
APIs: - AI Search Model: An API endpoint (REST or WebSocket) to send prompts and receive search results. - Web Crawler Service: An API endpoint for triggering web crawls based on provided parameters (e.g., target URLs, variables).
The PromptTemplateManager gives users the tools to
  1. Create prompts templates with variables from the current set of records loaded in the RecordCollector through Variable Mapping: mapping variables from current systems into the prompt templates before execution.
    1. See available variables from the RecordCollector, available via API.
    2. Enter a WYSIWYG editor for crafting custom prompts with syntax highlighting and auto-completion for supported languages (like Python, JavaScript). a MDX editor to write prompts.
      1. Variable Management: Interface to map variables from records available from the RecordCollector into the prompt template, possibly using a drag-and-drop or input field-based system.
    3. Upload markdown or mdx files.
    4. Save prompt templates that have valid variable syntax.
  2. Search, filter, and select saved prompts.
  3. Prompt Validation: Ensure prompts are syntactically correct and adhere to security guidelines.

3. Goals & Non-Goals

Goals

  • PromptTemplateManager uses a standard templating syntax to apply data from the RecordCollector to a PromptTemplate, resulting in a filled out template on a per-record basis.
  • Users can create, delete, and update PromptTemplates

Non-Goals

  • Does not send to AI Model APIs, nor evaluate responses.

Technical Design

Front-end Framework: React Back-end Framework: NEXT.js Container system: Docker
Pulls data from: RecordCollector Makes prompts available for RequestReviewer
Window to PromptTemplateManager available in MainContainerUI
sequenceDiagram participant User participant UI as PromptTemplateManager UI participant BE as Backend Service participant RC as RecordCollector participant DB as Template Database User->>UI: Create/Edit Template UI->>RC: Request Available Variables RC-->>UI: Return Variable List UI->>User: Show Variable Palette User->>UI: Design Template with Variables UI->>BE: Save Template BE->>DB: Store Template DB-->>BE: Confirm Save BE-->>UI: Save Confirmation User->>UI: Select Template & Record UI->>RC: Request Record Data RC-->>UI: Return Record Data UI->>BE: Request Template Population BE->>BE: Merge Variables BE-->>UI: Return Populated Template UI->>User: Show Preview

Components

ListColumn--Prompts ListItem--Prompt

Shared Components

Shared_Search Container Shared_Filter Container Shared_Filter Dropdown Shared_Prompt Template Editor Shared_MDX Editor SharedJsonEditor Shared_List-Column_Layout Shared_List-Row

Open Questions

How will we handle saving?

Comprehensive Wish List

Attribution of prompts to User Accounts. Filter prompts by team. Limit prompts to being visible and usable by team, lists of other users, and the organization (verified by email).
Source Control Management, Version Control
Error Collection: Collects errors associated with a specific prompt.