Implementation:Ucbepic Docetl AI Chat And Prompt Improvement

Knowledge Sources	DocETL
Domains	Prompt_Engineering, User_Interface
Last Updated	2026-02-08 01:40 GMT

Overview

Concrete UI components and backend endpoints for AI-assisted prompt iteration in the DocETL playground.

Description

The AIChatPanel React component provides a streaming chat interface using the ai/react useChat hook. The PromptImprovementDialog provides automated prompt rewriting with comparison rationale. The backend supports operation decomposition via WebSocket /ws/decompose/{client_id} which uses the FastDecomposer to split complex operations.

Usage

Open the AI Chat Panel from the playground sidebar to ask questions or request improvements. Use the decompose button on an operation card to automatically split it into simpler operations.

Code Reference

Source Location

Repository: docetl
File: website/src/components/AIChatPanel.tsx (L50-512), website/src/components/PromptImprovementDialog.tsx (L1-1270), server/app/routes/pipeline.py (L469-622)

Signature

// AI Chat Panel
function AIChatPanel(): JSX.Element
// Uses POST /api/chat with streaming response

// Prompt Improvement Dialog
function PromptImprovementDialog({
  operation, onUpdate, onClose
}: PromptImprovementDialogProps): JSX.Element

# Backend decomposition
@router.websocket("/ws/decompose/{client_id}")
async def websocket_decompose(websocket: WebSocket, client_id: str):
    """Decompose operation using FastDecomposer."""

Import

# Frontend: React component imports
# Backend: included via app.include_router(pipeline.router)

I/O Contract

Inputs

Name	Type	Required	Description
chat messages	string	Yes	User questions/instructions in natural language
operation prompt	string	Yes	Current operation prompt text for improvement
sample output	list[dict]	No	Pipeline output for context

Outputs

Name	Type	Description
chat responses	streaming text	AI-generated suggestions and explanations
improved prompts	string	Rewritten operation prompts
decomposed operations	Operation[]	Split operations from decomposition

Usage Examples

# Example AI Chat interaction
User: "Why are some entities being missed in the extraction?"
AI: "Looking at the output, I notice the prompt doesn't specify the
entity types clearly. Try updating the prompt to list specific entity
types: persons, organizations, locations, dates..."

Related Pages

Implements Principle

Principle:Ucbepic_Docetl_Prompt_Iteration

Requires Environment

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment